Skip to content

Release v1.1.0 - Design, memory and automated Missions - #1395

Draft
pascalandr wants to merge 264 commits into
mainfrom
dev
Draft

pascalandr wants to merge 264 commits into
mainfrom
dev

Conversation

@pascalandr

@pascalandr pascalandr commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

CodeNomad 1.1.0 - Design, memory and automated Missions

CodeNomad 1.1 brings image creation into a dedicated Design workspace, adds optional local agent memory, and lets Missions run on schedules and repository events. It also makes integrations easier to manage and long conversations easier to read and navigate.

Upgrade requirement: OpenCode 2.3.0 is now the minimum and recommended runtime. Updating CodeNomad does not automatically update or restart your shared OpenCode service. Review Settings → OpenCode before upgrading and explicitly restarting the service.

Design and Images

Optional persistent agent memory

  • Enable Memory to retain useful facts, preferences and references across conversations. Separate About you and Project scopes keep personal and project knowledge distinct; a repository's worktrees share project memory (feat(memory): add opt-in persistent agent memory #1313).
  • Review and edit entries in Settings, keep a local change history, and ask the agent to tidy a scope. Memory is local and off by default; disabling it retains saved entries and their history.
  • While Memory is disabled, its native tool stays out of the agent's tool catalogue. Enabling or disabling it updates running sessions without restarting OpenCode (fix(memory): hide the native memory tool while agent memory is disabled #1479).

Missions on schedules and repository events

Native integrations and Status

Clearer conversations and faster navigation

Desktop, remote access and reliability

Upgrade notes

Full changelog: v1.0.1...v1.1.0

pascalandr and others added 30 commits October 30, 2026 12:07
…andbox on native hosts (#1208)

Slice 6 (qa-native) of the Design image companion (SCR v1.3.1; AC-16,
AC-41, AC-52, AC-54). Test code only; no product source changes.

- `scripts/test-design-companion-native.mjs`: isolated OpenCode daemon,
scripted chat model, fake image model; AC-16 and native AC-52 (re-read
35 assistant messages later). Added to the PR build on
Windows/Ubuntu/macOS.
- `--ui <Design ZIP>` → `scripts/test-design-companion-ui.mjs` +
`fixtures/design-companion.tsx`: real UI S8 path with the installed
extension and real-transcript Markdown rendering (AC-52, AC-54).
- `scripts/panel-extension-native-probe.mjs`,
`test-panel-extension-native.mjs --api4` (also off Windows), new
`test-panel-extension-electron.mjs`: sandbox invariants with negative
controls (AC-41).
- `fixture-exemptions.ts`: registers the Images conversation and API 4
browser tests required by `fixture-wiring`.

Runs: Windows (OpenCode 2.2.1 native + UI path; WebView2 and Electron
runners), Ubuntu VM (WebKitGTK and Electron runners). **Not run:** macOS
WKWebView — the VM stops at its login screen.
…ct (#1209)

## Summary

- OpenCode **2.2.3** becomes both the **technical minimum** and the
**recommended** runtime. The minimum moves because upstream
[#60258](anomalyco/opencode#60258) (shipped in
2.2.2) changed the image tool contract that CodeNomad's Images gallery,
Design agent and Images settings consume: one name,
`experimental_image_generate`, for the tool, its permission action, its
job type and its completion notices' `metadata.source`, and the tool on
by default (only `experimental.image_generate: false` disables it). On
2.2.2+, the gallery lost every generation entry and the Design companion
fixture timed out; 2.2.1 passed.
- Images are implemented for the 2.2.3 contract only, with no fallback
to the old name: `IMAGE_TOOL` drives the gallery and its proof;
`imageGenerationSelection` reads unset as on; the Design agent and
skills name the new tool; the activation fixture exercises the image
tool instead of falling back to webfetch; the gallery fixture represents
the 2.2.3 transcript format.
- Older daemons (2.1.x, 2.2.1, 2.2.2) go through the existing
below-minimum path (`native_contract`). The 2.1.0 client/plugin/schema
pins stay.
- Help (all ten locales), READMEs, `CONTRIBUTING.md`, `dev-docs`
(compatibility, pruning, panel extensions, settings) and the
architecture skill describe the new minimum, its reason and default-on
image generation. The setup minimum-reason UI string is updated in all
locales.
## Contract review (v2.2.1 → v2.2.2 → v2.2.3; 40 + 101 commits)

Reviewed from the exact tags `v2.2.1` (`3ebcf9a32c`), `v2.2.2`
(`098aa7fd6e`) and `v2.2.3` (`082ac6281f`), the published npm packages,
and an authenticated OpenAPI dump from each official CLI running in its
own sandbox. The [changelog](https://opencode.ai/changelog) has entries
for both 2.2.2 (29 October, highlights only) and 2.2.3 (30 October, full
notes); everything below was also checked against the commit log and
source.

**Headline:** the HTTP contract CodeNomad consumes didn't change, but
2.2.2 **renamed the image tool** from `image_generate` to
`experimental_image_generate` and **turned it on by default**
([#60258](anomalyco/opencode#60258)).
CodeNomad's Images/Design features match the old name and the old
default, so they break on 2.2.2 and later (see [Design companion
finding](#design-companion-finding)). Nothing else found affects
CodeNomad.

- **Authenticated OpenAPI:** **134 paths and 162 operations** on all
three, the same operation set. Schemas 355 → 350 → 351.
  - **Added/removed operations:** none in either release.
- **Changed operations (2.2.2):** `GET /api/model`, `GET /api/event`,
`PUT /api/pty/{ptyID}`, `POST /api/shell`, `GET /api/shell/{id}/output`,
`POST /api/debug/heap-dump` and `GET /api/experimental/migration/v1`.
All of these are **representation-only**: the Effect 4.0.6 JSON Schema
output flattens nested `anyOf`, bounds integers with `maximum:
9007199254740991` (and `minimum` for signed integers), and rewrites the
numeric-string query pattern. The event-type and model-type enums are
unchanged.
- **Changed operations (2.2.3):** only the description of `POST
/api/credential`, which can now be refused by an `integration.connect`
policy.
- **Consumed schemas:** the same family selector as the 2.2.1 review
(`V2EventEncoded`, `Session.*`, `SessionStats.*`, `Form.*`,
`Permission.*`, `Prompt*`, `Event.*`, `Location.*`, PTY/Shell).
Re-derived from that description it selects 104 schemas at 2.2.1 (the
2.2.1 file says 103), all byte-identical from 2.1.4 to 2.2.1.
- In 2.2.2 the seven alias copies `Form.Fields_1..3` and
`Session.Message.Native_1..4` collapse into their base schemas, which
are byte-identical to the copies, leaving **97**. Of these, 71 are
byte-identical and 24 differ only in generator representation.
- Two changes are additive and optional: `Session.Info.time.busy`, the
start of the latest execution
([#59910](anomalyco/opencode#59910)), and the
input field `PromptInput.FileAttachment.source` (`Prompt.FileSource`,
inline or uri; CodeNomad doesn't send it).
- `V2EventEncoded`, `Session.Message.*` content, `Session.Inbox.*`,
`Form.*` semantics, `Permission.*` and `Event.SeqEncoded` are unchanged.
    - From 2.2.2 to 2.2.3 the family is byte-identical.
- **`Config.InfoEncoded`:** no semantic change in 2.2.2 (representation
only). 2.2.3 extends the managed-policy `action` enum (top-level and
`experimental.policies`) with `integration.connect`, `mcp.use`,
`skill.use` and `plugin.use`
([#60315](anomalyco/opencode#60315)); CodeNomad
writes no policies. The *meaning* of `experimental.image_generate`
changed: unset now means on, and only `false` turns it off. The schema
shape stayed the same and the OpenAPI carries no description, so no
contract signal reveals it.
- **Schedules (not consumed on dev):** `when` becomes optional;
`if`/`until` conditions (shell in 2.2.2, Code Mode in 2.2.3) are added;
outcomes gain `checked`/`expired`.
- **Admission:** every predicate in `compatibility/negotiate.ts`
(unchanged since the 2.2.1 review) holds on 2.2.1, 2.2.2 and 2.2.3.
- **Instance proxy:** no new routes and an unchanged allowlist. Proxied
routes whose representation changed: `GET /api/model`, `PUT
/api/pty/{id}`, `POST /api/shell` and `GET /api/shell/{id}/output`.
Integer bounds and an equivalent numeric pattern accept everything
CodeNomad sends.
- **Published declarations** (2.2.1 → 2.2.3):
- **Client:** 185 methods on all three; the method tree is identical.
2.2.2 adds the generated types `ScheduleCondition*`; `FormFields2` is
renamed to `FormFields1`, which is generated-type churn. 2.2.3 adds the
Promise media helpers `withMediaRefs`, `mediaHashFromUri` and
`MediaRequestOptions`
([#60116](anomalyco/opencode#60116)) and
persistent-PTY frame helpers in `solid/pty`. CodeNomad imports none of
these.
- **Schema:** 2.2.2 adds schedule conditions (`ShellCondition`,
`CheckMode`, …) and exports `SessionLegacy.part`. 2.2.3 adds
`CodeCondition*` and `policy.Action`.
- **Plugin:** 2.2.2 adds the GUI
`ComposerQuoteAnchor`/`ComposerFocusOptions`. 2.2.3 has no export
changes, but the optional peers `@opentui/*` become `>=0.8.0`
([#60385](anomalyco/opencode#60385)). The
published `dependencies.effect` moves from **4.0.1 to 4.0.6** (2.2.2).
- **Typecheck:** the server and UI typecheck with **0 errors** against
the 2.2.2 and 2.2.3 client/schema/plugin/effect declarations through
scratch `paths` overrides. The pinned 2.1.0 control is also clean. The
harness now also remaps `effect` itself; the 2.2.1 harness skipped it
because effect's `exports` are plain `.js` strings.
- **Host modules:** the three daemon-loaded bundles still import
`@opencode/plugin` (`Plugin`), `/effect` (`Plugin`, `Rpc`, `Location`),
`/promise/adapter` (`fromPromise`), `/rpc` (`Rpc`), `effect` and
`effect/encoding` (`Base64Url`) from the host. A probe plugin in an
isolated **2.2.3** daemon resolves every module and binding, as on
2.2.1. `runtime-modules.ts` is unchanged between the tags.
- **Host Effect: 4.0.1 → 4.0.6** (2.2.2,
[#59630](anomalyco/opencode#59630)). The 2.1.0
`@opencode/schema`/`@opencode/client` code bundled into the Missions and
pruning plugins now runs on host Effect 4.0.6. The plugin-runtime,
pruning, Missions-contract, permission-receipts and image-settings
native suites pass on 2.2.3.
- **Private native services:** the keys (`LocationServiceMap`, `Form`,
`Permission`, `Location`, `Job`, `Database`, `Session`, `Bus`) and the
members CodeNomad calls are unchanged. The changes are additive:
`Permission.Unattended` (a reference that denies unanswerable asks for
schedule checks), a `CorrectedError.message` getter, and `Session`
deciding fork media from the parent's location.
- **Storage** (three migrations): `session_part` (new table),
`session_v2.time_busy` and four `schedule` check columns.
- **`session_part`**
([#58915](anomalyco/opencode#58915)): history
reads now *prefer* `session_part` rows over
`session_message.data.content` for assistant messages. The projector
writes parts there only for messages that already have part rows, so new
messages stay inline. Every 2.2.3 fixture DB inspected has **0**
`session_part` rows, and upstream `dev` (`1e28374e48`, 30 October) has
no writer either.
  - The tables CodeNomad reads directly keep their shape.
- **Forward risk:** when a release starts writing parts, CodeNomad's
direct `session_message.data.content` readers and the pruning `UPDATE …
json_set(data,'$.content',…)` (`session-pruning/transaction.ts`) would
silently miss or be shadowed by part rows. See Recommendations.
- **Behavior changes checked:**
- **Image generation** (2.2.2): renamed and on by default. See the
[dedicated section](#design-companion-finding).
- **Code Mode:** inner tool progress no longer replaces the `execute`
call rows ([#60480](anomalyco/opencode#60480)),
and progress is throttled to at most one update per 100 ms
([#60120](anomalyco/opencode#60120)). Programs
see the same failure text as direct calls
([#60109](anomalyco/opencode#60109)), `fetch` in
Code Mode is now subject to `webfetch` permissions
([#59954](anomalyco/opencode#59954)), `for…in`
visits inherited keys
([#59933](anomalyco/opencode#59933)) and `null`
survives text encoding
([#57639](anomalyco/opencode#57639)). The
run-to-completion regression behind the 2.1.2 minimum passes on 2.2.3:
`{ total: 2000, beforeAwait: 1000, count: 1000 }`.
- **Permissions and MCP:** permission reasons and feedback are kept in
MCP tool failures
([#60102](anomalyco/opencode#60102)). Shell
permission checks moved fully off tree-sitter and now check redirects,
PowerShell method calls and dynamic directory changes
([#60332](anomalyco/opencode#60332),
[#60376](anomalyco/opencode#60376)), so some
shell commands may now ask where they didn't. The managed-policy split
(`integration.use` vs `mcp.use`/`skill.use`/`plugin.use`) is a changelog
"breaking change" for managed deployments only.
- **Sessions:** deleting a session removes its persistent terminals
([#60322](anomalyco/opencode#60322)), and a
persistent terminal with a missing cwd is rejected
([#60300](anomalyco/opencode#60300)). Forked and
imported sessions store media as blobs when `media_blob` is on
([#60118](anomalyco/opencode#60118)). The native
fork suite passes on 2.2.3.
- **Providers:** a built-in **Claude Agent** provider uses the user's
Claude Code login
([#58782](anomalyco/opencode#58782)).
Plugin-supplied model variants are preserved
([#60097](anomalyco/opencode#60097)), and the
dynamic provider npm fallback runs after external SDK hooks
([#60380](anomalyco/opencode#60380)). These are
source-reviewed and touch no consumed contract.
- **Persistent PTY binary** (the process daemon): 2.2.2 switches from
`@opencode-ai/pty` 0.3.1 to `@opencode/pty` **0.3.3**
([#60276](anomalyco/opencode#60276)).
`process-daemon-v0.3.3` contains only the release commit after 0.3.2 (a
version bump of the `@opencode/pty-*` binaries). **v0.4.0** (terminal
renaming, [#60324](anomalyco/opencode#60324)) is
**not** shipped in CLI 2.2.3, which pins 0.3.3. CodeNomad only lists
persistent PTYs (worktree deletion guard), whose schema is unchanged
apart from representation.
- **Client fix not inherited:** the 2.2.3 client evicts missing sessions
on 404 ([#60459](anomalyco/opencode#60459)).
CodeNomad stays on the 2.1.0 client.
- **`@opencode/ai` breaking changes** (media moved to `/experimental`,
one `tool()`): CodeNomad doesn't import `@opencode/ai`.
- Desktop/TUI-only changes (timeline performance, side-panel tabs,
spellcheck, question sounds, OpenTUI 0.8) don't affect CodeNomad.

**Verdict:**

- The **2.1.0 client/plugin pins stay**. Bumping them would bring
nothing CodeNomad consumes.
- No consumed HTTP/plugin contract broke; the breaking change is the
image tool contract of #60258, consumed by the Images gallery, the
Design agent and the Images settings.
- **Decision applied in this PR:** 2.2.3 becomes both the minimum and
the recommendation, and Images are implemented for the 2.2.3 contract
only (see Summary). This replaces the review's initial suggestion to
keep the 2.1.2 minimum and accept both tool names.

**Not verified here:**
- the full native suite matrix on 2.2.2/2.2.3; only the runs in Appendix
I were performed;
- the forward-compatibility of the `session_part` readers against a
runtime that writes parts (no such runtime exists yet);
- the Claude Agent provider and shell-permission changes at runtime
(source only);
- macOS/Linux.

---

## Design companion finding

**Symptom:** `scripts/test-design-companion-native.mjs <cli>` passes on
2.2.1 and fails on 2.2.2 and 2.2.3 with `Error: Timed out: series 1
generated and read`. Reproduced independently in a scratch checkout of
`origin/dev` (`f39642bc4`), with `CODENOMAD_NATIVE_FIXTURE_PARENT`
outside the home:

```
test-design-companion-native 2.2.1  exit=0  PASS AC-16, AC-52
test-design-companion-native 2.2.2  exit=1  Timed out: series 1 generated and read
test-design-companion-native 2.2.3  exit=1  Timed out: series 1 generated and read
```

**What the step waits for**
(`test-design-companion-native.mjs:238-241`): the fixture's loopback
chat provider answers a `RUN [...]` turn with two tool calls `{ tool:
"image_generate", args: { prompt, path: ".codenomad/design/icon/<name>"
} }`. The step then waits until the transcript holds two synthetic
notices with `metadata.source === "image_generate"` (line 209) and two
`read` parts of the generated files. No HTTP generate route, file watch
or media blob is involved; it is the agent tool path.

**Cause: an intentional upstream contract change, not an upstream bug.**
In 2.2.2, [#60258](anomalyco/opencode#60258)
(`d4e48f54da`, "turn experimental_image_generate on by default") changed
`packages/core/src/tool/plugin/image-generate.ts`:

- `export const name = "image_generate"` →
`"experimental_image_generate"`. That one constant is the tool name, the
permission action (`assertions: ["experimental_image_generate", "edit",
"read"]`), the Job `type`, and the completion notice's
`metadata.source`. `session/execution/restart.ts` also tags
restart-recovery notices `source: "experimental_image_generate"`.
- The opt-in became default-on: `if (!enabled || imageGenerate(entries)
=== false) return`. An unset `experimental.image_generate` now offers
the tool; only `false` removes it ("still honored for configs from when
the tool was opt-in; permissions are the supported way to turn it off").
- The 2.2.2 changelog states it: "agents can now use
`experimental_image_generate` without additional configuration".

2.2.2 also wraps the notice text as `<image id="…"
state="…">\n…\n</image>` and adds `id` to the tool output
([#60098](anomalyco/opencode#60098)).
CodeNomad's `Saved image: (.+) \(` parse still matches; `jobID`, `state`
and `path` metadata are unchanged.

**Evidence** (fixture DBs copied read-only to `evidence/`):

```
2.2.3, original fixture (calls "image_generate"):
  seq 5  tool image_generate  error  "No tool named \"image_generate\" is currently available. Please use a tool from the available tool list."  (x2)
  -> no image job, no notice, no read: the wait times out.

2.2.3, fixture with the tool name changed (scratch copy; CodeNomad server code unmodified):
  seq 5  tool experimental_image_generate completed {path:…/icon-minimal-a, model:"fiximg/image", jobID:"image_18cfb59…"}
  seq 18 synthetic {"source":"experimental_image_generate","jobID":"image_18cfb59…","state":"completed",…}
         <image id="image_18cfb59…" state="completed"> | Image request finished. Saved image: …
  seq 20 tool read completed (x2) … seq 40/48/49 refinement
  -> files, image calls and reads all correct, then
     AssertionError: generations.length 0 !== 3   (test line 260: gallery "generation" entries)
```

So the product bug is real and independent of the fixture: on 2.2.2+
CodeNomad's Images gallery finds **no generation entries**. Images still
appear through their `read` parts, but without prompt, model, lineage or
"proven" origin, and `galleryImageReference` refuses them as `changed`.

**CodeNomad code that assumes the 2.2.1 name or default** (origin/dev):

| Where | Assumption | Effect on 2.2.2+ |
| --- | --- | --- |
| `session-pruning/assets-gallery.ts:56,139` | `part.name ===
"image_generate"` | no generation entries; references refused `changed`
|
| `session-pruning/assets-gallery-proof.ts:104,130` | tool name and
notice `source === "image_generate"` | no calls or notices, so no
provenance proof |
| `opencode/image-settings.ts` `imageGenerationSelection(undefined) →
false` (doc: "Reads … as OpenCode does") | unset means off | Settings →
Images shows **Off** and `images.status()` returns `"disabled"` while
the tool is offered; help step "Turn on image generation"
(`docs/help/en/extensions.md:73`) is no longer needed |
| `opencode/design/opencode/agents/design.md`, `skills/design-core`,
`design-logo`, `design-ui-mockup` | the agent is told to call
`image_generate` | the model sees only `experimental_image_generate`;
the first call can fail with "No tool named …" |
| `scripts/test-design-companion-native.mjs:125,209,238,301` | fixture
calls/filters `image_generate` | red on 2.2.2+ |
| `scripts/test-design-activation-native.mjs:76,307` | ask rule on
action `image_generate`; tool detection by that name | **passes but
loses coverage**: `askTool` falls back to `webfetch` on 2.2.2/2.2.3
(logged) |
| `dev-docs/SESSION_PRUNING_RPC.md`, `PANEL_EXTENSIONS.md`,
`SETTINGS.md` | describe `image_generate` parts and opt-in | out of date
for 2.2.2+ |

User permission rules written for action `image_generate` no longer
match on 2.2.2+; that is an OpenCode-side change for users. CodeNomad's
help never names the permission action; the Design fixtures and agent
use `experimental_image_generate`.

**Fix in this PR** (2.2.3 contract only, no fallback):

1. One `IMAGE_TOOL = "experimental_image_generate"` constant drives the
gallery and its proof; the gallery reads the 2.2.3 notice format.
2. The native fixtures use the new tool; the activation fixture
exercises it (default on) instead of falling back to webfetch.
3. The Design agent and skills name `experimental_image_generate`.
4. `imageGenerationSelection` reads an unset
`experimental.image_generate` as on; only `false` disables it. Settings
→ Images and `images.status()` report it that way.

## Follow-up

- Hardening for `session_part` (independent of this upgrade, no 2.2.3
regression): make the direct readers and the pruning transaction fail
safe when `session_part` has rows for the target messages.

## Isolated Windows qualification

Base: dev `f39642bc4`; Node 25.2.1 (local; CI runs the pinned 24.20.0);
exact official CLI **2.2.3**; locked client/plugin **2.1.0**;
`CODENOMAD_NATIVE_FIXTURE_PARENT` outside the user profile for the
Design fixtures.

- **Before the image fix (recommendation-only change):** every native
suite passed on 2.2.3 except the Design companion end to end, which
passes on 2.2.1 and times out on 2.2.2/2.2.3 at "series 1 generated and
read" (`No tool named "image_generate" is currently available`).
- **After this change, on 2.2.3:** Design companion (3 files, 2 series,
all proven, lineage), Design activation (image tool offered to Design
and Build, default on), Design content, image settings (default on,
Off→On hot reload), session pruning including session images/media, Code
Mode jobs `{ total: 2000, beforeAwait: 1000, count: 1000 }`.
- **Full native set on 2.2.3 (before the image fix, unaffected areas):**
plugin runtime, automation, Location/worktrees/event relay, permission
receipts, Git init/degraded, Missions (contract, family control, native,
trajectory) and all 11 recurring journeys, sessions (environment, fork,
aside, interrupt, lifecycle, blank cleanup, compaction, prompt skills,
providers, service usage, web search, plugin packages, output limit,
Promise-tool failure, subagent depth, temporary workspaces), migrations
from 2.1.2/2.2.1/2.2.2, install/upgrade/legacy replacement.
- **Units and checks after the change:** root and server typecheck
clean; server **2,825 passed, 9 skipped, 1 cancelled** (`index.test.ts`
shutdown test timed out under full-suite load; **4/4** alone); gallery
107/107; Design 23/23; opencode-update 62 passed / 1 skipped; touched UI
units 8/8; browser `opencode-setup` + `websearch-settings` 38/38; native
plugin build 2/2; ten-locale help 31/31.
…er the same name (#1210)

A failed image generation (here three Gemini calls rejected with 'API
key not valid') followed by a retry under the same file names made every
retried image 'origin not verifiable' in the Design light table, and the
retry's generation showed 'No read-back image attached': the proof saw
two possible writers of each file.

OpenCode 2.2.3 publishes a job's notice only after `jobs.wait`
(`image-generate.ts` `notifyWhenDone`), and a failed or cancelled job
saves nothing afterwards. `proveWriter` now leaves out a call whose
single decoded notice (per `generationState`) reports `error` or
`cancelled` before the remaining writer's start. Notified after the
retry started, never notified, or notified twice → still unverifiable.

- `assets-gallery-proof.ts` (+10/−1), `assets-gallery-bounds.test.ts` (5
cases × plain/blob; positives red without the change),
`dev-docs/SESSION_PRUNING_RPC.md`.
- Checks: session-pruning, session-pruning route and API 4 tests; server
typecheck.
Help tables broke words in the middle of narrow columns ("Compone nt", "CodeNom ad project tab" in Missions → Lifetime and service restarts). Cells used `overflow-wrap: anywhere`, which also lowers each cell's minimum width to one character, so the automatic table layout squeezed short label columns in favor of long description columns.

Cells now use `overflow-wrap: break-word`: an over-long word still wraps, but column widths are computed from whole words. Inline code keeps `overflow-wrap: anywhere`, and wide tables keep scrolling locally.

test-help-browser.mjs checks, in every English article at 390 and 1440 px, that no word outside code spans two lines in a table cell; it fails with the previous rule and the help browser suite passes (47/47). Review checked all ten locales at both widths.
… login (#1214)

Usage showed nothing for OpenCode's `claude-agent` provider (Claude Agent SDK), which signs in with Claude Code's own login (`claude auth login`): CodeNomad had no usage reader for that ID.

`claude-agent` is now an alias of the existing `claude-code` reader, which reads Claude Code's login on the daemon host; both IDs share one cached reading and backoff, and the login never appears in responses or logs. Without a Claude Code login it shows not configured, with no fallback to another provider's credential. PROVIDER_ACCOUNTS.md, the server README and the providers help page (ten locales) are updated.

Validation: service and route tests cover both IDs and fail without the alias; usage, route and credential tests (29/29), root typecheck and the help checks pass.
)

The Mission editor's Playbook selector showed only names; what each playbook does was documented only in the help. The selected playbook's one-sentence description now appears directly under the selector (referenced by its aria-describedby) and follows the selection.

Five `missions.control.template.<id>.description` strings reuse each locale's help wording in all ten locales; the paragraph takes the editor's existing secondary text style. The Missions help (ten locales) and dev-docs/MISSIONS.md mention it.

Validation: a browser test opens the real editor, checks the default description, its placement and fit, then Exploration and Compare two models; it fails without the component change. A unit test checks every locale has its own description for each playbook. Mission editor and automation browser tests (98/98), UI unit tests, help checks and the root typecheck pass.
A user asked whether sessions and sub-agents in a mission can use earlier sub-agent sessions and mission context. The Missions help now explains how context flows: the mission map as shared, durable memory; what the coordinator, native-subagent tasks, independent-session tasks and their helpers each receive; how results come back; how later tasks build on earlier ones (dependencies, reused conversations, existing independent conversations) and what compaction means for model context versus stored history; daily runs and follow-up missions; and on-demand search of earlier conversations through CodeNomad's session search tool.

The section (`#context-flow`, `#context-search`) is in all ten locales with each locale's UI labels; every statement is mapped to the code in the PR description.

Validation: help structure and browser checks pass (31/31, 47/47); two review rounds, the second with no findings.
…t now (#1219)

A user reported that pausing a mission stopped it abruptly, mid-step; pausing by hand ("pause this session" queued to the coordinator and sub-agent) let the current step finish and resumed cleanly. Pause interrupted the whole native family at once.

Pause from the panel is now graceful (`graceful: true`): it records a paused control with no targets and no native side effect, so it settles at once. The paused context every actor reads at each model request asks it to finish only the step in progress, start no new work or sub-agents and end its turn; the plugin's context hook adds it for native descendants, walking ordinary sub-agents' ancestry (bounded, cached parent IDs) to their actor and re-reading a paused chain so only sessions still placed with the actor's admitted project and Location receive it. Dispatch blocking and report parking apply at once, as before.

While the activity projection observes a running, queued, background, Form or permission actor of a gracefully paused mission, the panel shows "Pausing: the steps in progress are finishing." and Interrupt now, which sends the existing immediate Pause (family interruption and receipts unchanged). An immediate Pause is valid from paused only after a graceful one. Play after a graceful Pause tells the coordinator that delegated work may have stopped early and should be continued, not restarted. The flag is carried by the route, RPC schema, plugin, journal (graceful means no targets), projection and UI intents; requests without it keep the immediate behavior.

MISSIONS.md, the Missions help (ten locales) and three UI strings (ten locales) are updated. Tests cover the lifecycle transitions and replay, the journal rule, the real plugin context hook for the coordinator and three levels of sub-agents (including moved sessions and warm caches), and the real panel for running and Form-waiting work. Three review rounds; the last had no findings.
…th their names (#1218)

The Model selectors of mission profiles (Settings → Missions defaults, Mission editor Options, Compare Model A/B) listed every catalog model as raw `provider/id`, unfiltered. They now show the same list and names as the conversation's model selector.

The composer's catalog building moves unchanged into `lib/model-catalog.ts` (`buildProviderCatalog`, `listCatalogModels`), and preference filtering into `stores/model-picker.ts` (`preferredModelList`): hidden models are left out, favorites-only applies while a favorite remains, and models sort by provider then name. Mission controls reuse them, keep only tool-capable models, and group options by provider. Summaries show friendly names without their own catalog read (`knownModelName`), with names keyed by the workspace-root Location and connection generation so a fresher conversation catalog wins. A saved value the list leaves out stays selected under its name, or as "· Unavailable" when the selector cannot offer it; stored values keep the `providerID/id` format. A global `optgroup` rule styles provider groups in native selects.

Help (ten locales) and MISSIONS.md describe the list. Tests cover the shared helpers, freshness and Location fences, and the real Settings, editor and Compare selectors, including hidden, favorite-filtered, unknown and tool-less saved values; three review rounds, the last with no findings. Rebased on #1219: typecheck, help checks and 77 mission/model browser tests pass.
Setting many rows of the same kind (Message content eyes, session
selection checkboxes, model visibility) took one click per row. Image
editors let users drag across a column of layer eyes instead; CodeNomad
now does the same.

Pressing a toggle and dragging out of it flips it, and every toggle of
the same column or list the pointer then crosses takes that new state;
toggles already in that state stay as they are. Releasing the button
clicks nothing, including the first toggle when the drag returns to it.
It works for mouse and pen; touch drags keep scrolling, and a touch
landing during a mouse drag leaves the sweep going.

Implementation: `lib/toggle-sweep.ts` installs one capture listener on
first use. Toggles opt in with `data-toggle-sweep={toggleSweepGroup(...)}`,
one group per column or list, so the Message content eyes and chevrons
sweep separately and exclusive selectors (tabs, presets) never sweep.
Each toggle changes through its own click, so its handlers, disabled
state and accessibility are unchanged; state is read from `checked`
(labels stand for their checkbox) or `aria-pressed`/`aria-checked`. The
drag is sampled every 4 px so fast moves skip no row, text selection
and native drag are blocked during a sweep, the trailing click is
swallowed, and a sweep whose release went elsewhere ends at the next
move without the button. `afterSweep` defers costly work until the
sweep ends.

Groups: Message content visibility and expansion columns, session
selection checkboxes, right-panel customization (built-in tabs and
Status sections; extension activation is a trust decision and stays
out), notification routing, model visibility, and the multi-select
options of agent forms and provider sign-in forms.

Message content dropped clicks made while a save was pending and showed
a row only once the server confirmed it, re-rendering the transcript
for each. Rows now show their requested state at once; a sweep saves
once, on release, so the transcript re-renders once. Saves run in
order and each batch of rows is sealed as it joins the queue, so a
later click never jumps ahead of a preset change nor rides along with a
sweep still held.

Notification routing shows each box at once, writes the whole table
once per sweep, runs its writes one after another so an older table
never lands last, and asks for System permission once per gesture.
Model visibility already showed pending changes and wrote them in
order.

Validation: browser regressions in transcript-presets (a single jump
across four eyes with slow saves, shown during the drag with nothing
saved, then one write on release; a hidden row that stays hidden; a
release back on the origin; the chevron column alone; a held sweep not
persisted by queued saves; a row clicked after a preset applying after
it), session-search (both directions, a release over a row title
opening no session, a lost release leaving no sweep, a touch mid-drag)
and notification-routing (one write per sweep, serialized writes with
the last winning, one permission request). Each fails with its part of
the fix removed. Two gatekeeper rounds ended with no P0-P2 finding; the
user validated the behaviour in a separate Electron dev instance.
dev-docs/UI_CONVENTIONS.md and dev-docs/TRANSCRIPT.md record the
convention, and the help describes the gesture in all ten languages.
Follow-up to #1183, which merged with the sticky creation footer. That
footer's second review left two P2s open:
- **R3:** a stale-map or cleanup-unavailable banner rendered after the
form moved Create by about 32 px during a native automation click.
- **R4:** on short panels, the sticky footer covered the focused field
or made Options unreachable.

## Change
- **Layout:** while creating, the form fills the Missions panel as a
column. Its **body is the only scroller**, and the footer sits below it,
outside that scrollport, so native focus scrolling keeps the focused
field visible.
- **Banners** shown during creation are ordered above the form. They
shrink the body and never move the footer; their text and roles are
unchanged.
- **Short panels** (the footer plus an 8rem body don't fit): the whole
form scrolls in the panel, with the footer after it. This uses the
existing short-layout helper, renamed `mission-panel-scroll.ts` and
shared with the task tree, which re-reveals the focused element when the
mode switches.
- **Unaffected:** edit/delete forms and the Preferences depth editor.

## Validation
- **automation-forms:** five held outcomes (three depth reads, a stale
banner, cleanup unavailable) released during the real controller's
click. Create's box is unchanged and exactly one POST is sent.
- **mission-panel-host:**
- A real-keyboard Tab pass at 1000x600 and 260x300, with Options and the
brief prompt open, hit-tests every focused control.
  - At 220x120, Options opens and every control is reachable.
- **Red checks:** the sticky CSS fails both banner cases (-32.5 px) and
the focus checks; removing the fallback fails the short-panel checks.
- **Passing on current dev:** UI typecheck; automation-forms (2 runs),
mission-panel-host, mission-control, mission-panel-layout,
mission-panel-geometry, mission-create-submit, mission-create-refusals,
mission-editor-lifetime.
notification-routing.test.ts (#1221) serves the status-tokens fixture through startProductFixture like its sibling tests but was missing from the fixture exemption map, so fixture-wiring.test.ts failed browser shard 1. It now carries the same exemption. fixture-wiring.test.ts and notification-routing.test.ts pass locally (3/3).
## Summary

- OpenCode **2.2.5** becomes the **recommended** (qualified) runtime.
The **2.2.3 minimum** and the 2.1.0 `@opencode/client`/`plugin`/`schema`
pins stay: neither 2.2.4 nor 2.2.5 changes a contract CodeNomad
consumes.
- `RECOMMENDED_OPENCODE_VERSION` is 2.2.5, and the reviewed-publication
shortcut in `contractProfile` lists 2.2.3 and 2.2.5; other versions,
2.2.4 included, keep negotiating their authenticated schema. The CI
`latest` lane already resolves the latest stable CLI.
- Help (all ten locales), READMEs, `CONTRIBUTING.md`,
`dev-docs/OPENCODE_COMPATIBILITY.md` and the architecture skill state
two separate versions again: minimum 2.2.3 with its reason,
recommendation 2.2.5.

## Contract review (v2.2.3 → v2.2.4 → v2.2.5; 31 + 19 commits)

Reviewed from the exact tags `v2.2.3` (`082ac6281f`), `v2.2.4`
(`45898d0f85`) and `v2.2.5` (`9bbb8ce18c`), the published npm packages,
and an authenticated OpenAPI dump from each official CLI running in its
own sandbox. The [changelog](https://opencode.ai/changelog) has full
notes for 2.2.4 and 2.2.5 (both 30 October); everything below was also
checked against the commit log and source. CodeNomad side: `origin/dev`
`605820669`.

**Headline:** neither release changes a contract CodeNomad consumes.
2.2.4 adds opt-in **artifacts** (seven `/api/experimental/artifacts`
operations, four tools, a plugin hook), **renamable persistent
terminals**, and moves background-job recovery markers from `kv` into a
new **`job` table**. 2.2.5 is a bug-fix release; its one schema change
is that `compaction.buffer` accepts a percentage such as `"20%"`. The
image tool contract adopted in #1209 (`experimental_image_generate`, on
by default) is unchanged through 2.2.5, and still nothing writes
`session_part`.

- **Authenticated OpenAPI:** 134 → **138** → 138 paths, 162 → **169** →
169 operations, schemas 351 → 362 → 363.
- **Added operations (2.2.4):** `GET
/api/experimental/artifacts/providers`, `GET|POST
/api/experimental/artifacts`, `GET|PATCH|DELETE
/api/experimental/artifacts/{artifactID}`, `GET
/api/experimental/artifacts/{artifactID}/view/*`
([#60351](anomalyco/opencode#60351)). None
removed in either release.
- **Changed operations (2.2.4):** `GET /api/event`, where the `types`
filter enum gains `persistent-pty.updated` and `artifact.updated` (94 →
96 types, none removed). `PUT /api/experimental/persistent-pty/{ptyID}`
can also rename (`title`; `null` clears it), `size` becomes optional,
and a 501 `PtyFeatureUnsupportedError` is returned while an older
terminal service can't rename
([#60329](anomalyco/opencode#60329)).
  - **Changed operations (2.2.5):** none.
- **Consumed schemas:** same family selector as the 2.2.3 review, 97
schemas at every version.
- 2.2.3 → 2.2.4: 95 byte-identical. The two that change,
`PersistentPty.Info` (+ `userTitle`, nullable, required) and
`PersistentPty.UpdateInput` (`size` optional, + `title`), are both
additive.
    - 2.2.4 → 2.2.5: all 97 byte-identical.
- `V2EventEncoded`, `Session.*`, `Form.*`, `Permission.*`, `Prompt*`,
`Event.*`, `Location.*`, PTY/Shell are byte-identical from 2.2.3 to
2.2.5.
- **`Config.InfoEncoded`:** 2.2.4 adds `experimental.artifacts`
(boolean, opt-in). 2.2.5 turns `compaction.buffer` into
`Config.Compaction.Percent | integer`
([#60647](anomalyco/opencode#60647)).
`experimental.image_generate`, `experimental.media_blob` and
`experimental.schedules` are unchanged.
- **Admission:** every predicate in `compatibility/negotiate.ts`
(unchanged on origin/dev) holds on 2.2.3, 2.2.4 and 2.2.5.
`contractProfile` lists only `2.2.3` as reviewed, so 2.2.4/2.2.5 are
admitted through authenticated contract recognition.
- **Instance proxy:** the allowlist is unchanged on origin/dev. No added
or changed route is proxied: artifacts and persistent PTYs are outside
it. The proxied `GET /api/config` returns the additive config fields
above.
- **Published declarations** (2.2.3 → 2.2.5):
- **Client:** 185 → **192** → 192 methods. 2.2.4 adds
`experimental.artifacts.{providers,list,create,get,update,delete,view}`,
the artifact and `PersistentPtyUpdated` types, and
`PtyFeatureUnsupportedError`. It removes
`createPtyClient`/`PtyConnectInput` from `@opencode/client/solid/pty`,
which CodeNomad doesn't import. 2.2.5 adds `ConfigCompactionPercent`.
- **Schema:** 2.2.4 adds the `artifact` module and
`PersistentPty.Updated`; 2.2.5 adds `config/compaction` `Percent`.
- **Plugin:** 2.2.4 adds `effect/artifact` and `promise/artifact` and
`ctx.experimental.artifact` (`transform`, `reload`). 2.2.5 has no export
changes.
- **Typecheck:** the server and UI typecheck with **0 errors** against
the 2.2.4 and 2.2.5 client/schema/plugin/effect declarations. The pinned
2.1.0 control is also clean.
- **Host modules:** the bundles' host imports are identical to the 2.2.3
review. A probe plugin in an isolated **2.2.4** and **2.2.5** daemon
resolves every module and binding. `runtime-modules.ts` is unchanged
from v2.2.3 to v2.2.5.
- **Host Effect:** stays **4.0.6** (`bun.lock` sha512 identical;
published `dependencies.effect` 4.0.6).
- **Persistent PTY binary:** the 2.2.4 and 2.2.5 CLIs ship
`@opencode/pty` **0.4.0** (string `opencode-pty 0.4.0` in the Windows
binary; 2.2.3: `0.3.3`). `process-daemon-v0.4.0` adds only terminal
renaming ([#60324](anomalyco/opencode#60324)):
an additive `rename` request and `features: ["rename"]` with an
unchanged `PROTOCOL_VERSION`. CodeNomad only lists persistent PTYs
(worktree deletion guard) and reads `status`/`cwd`.
- **Private native services:** the keys (`LocationServiceMap`, `Form`,
`Permission`, `Location`, `Job`, `Database`, `Session`, `Bus`, `Shell`)
and the members CodeNomad calls are unchanged. The changes are additive:
`Job.cancelSession` (2.2.4), `Permission.allows` (2.2.5; never asks,
used by `edit` to decide whether to quote file lines), and
`Session.remove` now calls `jobs.cancelSession`. The Missions recurrence
clock calls `Job.get/start/cancel`. Its Job metadata carries no
`sessionID`, so deleting a session doesn't cancel it.
- **Storage:** one migration, `20261030054400_job` (2.2.4,
[#60491](anomalyco/opencode#60491)). It creates
`job`, copies every `kv` row `job.background/%` into it and deletes
those `kv` rows. 2.2.5 adds none.
- CodeNomad reads only its own `kv` keys (Missions, pruning receipts and
claims), never `job.background/*`.
- Native migration 2.2.3 → 2.2.4 passes. Every 2.2.4/2.2.5 fixture DB
inspected holds the `job` table, **0** `job.background/` rows and no
leftover `schedule:leader` lease.
- **`session_part`:** still **not written**. `history.ts`/`projector.ts`
are unchanged from v2.2.3 to v2.2.5, the projector still only rewrites
existing part rows, and every inspected 2.2.4/2.2.5 fixture DB has **0**
rows. The forward risk from the 2.2.3 review stands (Recommendations).
- **Behavior changes checked** (plain-language list in [What users will
notice](#what-users-will-notice)):
- **Image tool** (#1209 contract): `image-generate.ts` is unchanged from
v2.2.3 to v2.2.5. 2.2.4 refactors restart recovery into one
`restartMessage` but keeps `metadata.source:
"experimental_image_generate"`, `jobID`, `state`, `path` and the `<image
id state>` notice text. Design companion, activation, content and image
settings pass natively on 2.2.4.
- **Schedules** (2.2.4,
[#60478](anomalyco/opencode#60478), then
`da87c475a7`): the loop now runs in every server process, including
`opencode serve`, and elects a leader through a `kv` lease
`schedule:leader` (claimed at boot, released when there are no active
schedules). Firing stays opt-in (`experimental.schedules === true`).
CodeNomad uses neither OpenCode schedules nor that key; Missions
recurrence uses its own storage and the Job service.
- **Sessions:** deleting a session now cancels its running background
jobs (shell, image, subagent) and drops their markers (2.2.4). That is
the intended upstream cleanup; CodeNomad's session deletions
(blank-session cleanup, temporary workspaces) get it automatically.
- **Events:** `persistent-pty.updated` and `artifact.updated` reach
CodeNomad's shared event relay. The relay routes generically
(`instance-events.ts`), the 2.1.0 client's SSE reader doesn't validate
types, and the UI handles neither type. The native relay suite passes on
2.2.4.
- **Tools** (2.2.5): a model call to `bash` runs as `shell` with shell
permissions, and the result carries a note
([#60687](anomalyco/opencode#60687)). CodeNomad
already groups and renders `bash` and `shell` alike
(`message-part-grouping.ts:48`, `tool-call/utils.ts`). `read`/`edit`
failure texts are richer
([#60688](anomalyco/opencode#60688),
[#60689](anomalyco/opencode#60689)).
- **Compaction** (2.2.5): the prompt is capped at the context window
when a catalog's input limit is larger
([#60660](anomalyco/opencode#60660)), and
`compaction.buffer` may be `"N%"`. CodeNomad never reads or writes
`compaction.buffer`. LiteLLM context errors now trigger compaction
([#60692](anomalyco/opencode#60692)).
- **Reasoning replay** (2.2.5,
[#60498](anomalyco/opencode#60498)): reasoning
is no longer re-sent as plain text after a model switch, and interrupted
or errored reasoning is dropped from the request. This changes provider
requests only; stored transcripts and `Session.Message.*` are
byte-identical.
- **MCP** (2.2.5): stdio messages over 10 MiB no longer disconnect a
server ([#60694](anomalyco/opencode#60694)), and
an idle Location no longer restarts its MCP servers right after shutting
down ([#60668](anomalyco/opencode#60668)).
- **Providers:** a Console-supplied provider can now be overridden by
user, project and remote config (2.2.4). The Claude Agent provider only
replaces Claude Code's `/login` hint with `claude auth login` in its
errors ([#60592](anomalyco/opencode#60592)); the
guided setup
([#60603](anomalyco/opencode#60603)) is OpenCode
Desktop only. The provider ID (`Provider.ID.claudeAgent`) and its use of
Claude Code's own login are unchanged, so CodeNomad's `claude-agent`
quota reader (#1214) is unaffected.
- **Service lifecycle:** the CLI now adds `127.0.0.1,localhost,::1` to
`NO_PROXY`/`no_proxy` for itself and its children
([#60441](anomalyco/opencode#60441)), so
loopback traffic from daemon-hosted code, CodeNomad's plugins included,
bypasses environment proxies. CodeNomad sets no proxy variables.
- **Code Mode, Forms, permission prompts, media blobs:** no source
changes from v2.2.3 to v2.2.5 (`packages/codemode` only bumps its
version). The Code Mode run-to-completion regression passes on 2.2.4: `{
total: 2000, beforeAwait: 1000, count: 1000 }`.
- **`@opencode/ai`:** one `ai` object for Promise and Effect
([#60516](anomalyco/opencode#60516); left out of
the changelog by
[#60631](anomalyco/opencode#60631)). CodeNomad
doesn't import `@opencode/ai`.
- Desktop/TUI-only changes (desktop terminals on persistent PTYs,
base-branch and worktree shortcuts, quote sending, flat inputs, palette
ordering) don't affect CodeNomad. One indirect effect follows.
- **Worktree deletion:** OpenCode Desktop terminals now live as
persistent PTYs that survive restarts
([#60346](anomalyco/opencode#60346)).
CodeNomad's guard (`routes/worktrees.ts`, "Running persistent PTY blocks
deletion: <id>") will therefore refuse more often for users who also
open terminals in OpenCode Desktop. The behavior is correct, but the
message names the PTY ID rather than the terminal's title.

**Verdict:**

- The **2.1.0 client/plugin pins can stay**. Bumping them would bring
only artifacts, PTY renaming and a config type, none of which CodeNomad
consumes.
- The **2.2.3 minimum has no reason to move**. No consumed HTTP, plugin,
private-service or storage contract broke in 2.2.4 or 2.2.5.
- **Recommend 2.2.5.** It is admitted today through contract
recognition. Recommending it is CodeNomad-side bookkeeping:
`RECOMMENDED_OPENCODE_VERSION` (`runtime-support.ts`), the
reviewed-version check in `contractProfile`
(`compatibility/runtime.ts`), the version texts in docs and help, and
CI's qualification target. This is subject to the full native suite on
the exact 2.2.5 CLI, run separately.
- **No CodeNomad bug found** in 2.2.3 → 2.2.5. Risks: the `session_part`
forward risk (unchanged) and the more frequent worktree-deletion refusal
(UX).

**Not verified here:**
- the full native suite on 2.2.5: only a host-module probe ran natively
on 2.2.5, and everything else for 2.2.5 is source and OpenAPI review
(Appendix I);
- the 2.2.4 recurring Missions journey (`test-recurring-simple-native`
ABCDGELFWQN), which was interrupted by a server restart;
- the `kv` → `job` migration with a real pending background marker: only
the empty case ran, and upstream `job.test.ts` covers the copy;
- whether a `bash` call is stored as `bash` or `shell` (source suggests
the model's name is kept; CodeNomad renders both);
- artifacts, Claude Agent and MCP stdio framing at runtime (source
only);
- macOS/Linux.

---

## What users will notice

**OpenCode 2.2.4**
- **Persistent desktop terminals:** OpenCode Desktop terminals belong to
the session, keep their scrollback across server restarts, stay in sync
with the TUI and other windows, and can be renamed. Renamed terminals
keep their names.
- **Artifacts (opt-in):** with `"experimental": { "artifacts": true }`,
agents can publish, update and serve HTML or static artifacts
(`artifact_create`, `artifact_read`, `artifact_update`,
`artifact_delete`) through a built-in local provider, which stores files
under OpenCode's data folder, or a plugin-provided one.
- **Schedules:** enabled schedules also fire under `opencode serve`, and
only once when several OpenCode processes share a database. They stay
off unless `experimental.schedules` is `true`.
- **Claude Agent:** OpenCode Desktop guides installation, `claude auth
login` and connection checking. Error messages now say `claude auth
login` instead of `/login`.
- **Config:** user, project and remote configuration can override
Console-supplied provider names, limits, variants, settings and headers.
- **Networking:** requests to the local OpenCode service no longer go
through `HTTP(S)_PROXY`.
- **Sessions:** deleting a session also stops its running background
work, such as shells, image generations and subagents.
- **Desktop/TUI:** base-branch shortcut, simpler location popover,
faster Markdown streaming and various fixes.

**OpenCode 2.2.5**
- **Compaction:** `"compaction": { "buffer": "20%" }` keeps a share of
each model's context free; token counts still work. Compaction no longer
overfills models whose catalog input limit exceeds their context window,
and LiteLLM context errors trigger compaction instead of stopping.
- **Tools:** a model calling `bash` gets `shell` (with shell
permissions) instead of an error. `read` names the missing folder, and
`edit` shows the closest lines when `oldString` isn't found.
- **Reasoning:** switching models or resuming an interrupted reply no
longer resends reasoning as plain text, which some providers rejected.
- **MCP:** large stdio messages (over 10 MiB) no longer disconnect a
server, and idle locations don't immediately restart their MCP servers.
- **Desktop/TUI:** worktree and base-branch shortcuts, a shell-mode
toggle, sending a single quote from its popup, and MCP servers removed
from the connect menu.

## Follow-ups (outside this PR)

- Hardening for `session_part` (independent of this upgrade; 2.2.5 still
writes no rows): make the direct `session_message.data.content` readers
and the pruning transaction fail safe when `session_part` has rows for
the target messages.
- Optional UX: when the worktree deletion guard finds a running
persistent PTY, name it by `userTitle ?? title` (2.2.4+) and say it may
be an OpenCode Desktop terminal.
- Any future reader of `compaction.buffer` must accept `"N%"` as well as
a token count.

## Isolated Windows qualification

Base: dev `605820669`; Node 25.2.1 (local; CI runs the pinned 24.20.0);
exact official CLI **2.2.5**; locked client/plugin **2.1.0**;
`CODENOMAD_NATIVE_FIXTURE_PARENT` outside the user profile.

- **Native suites on 2.2.5, all passing:** Code Mode jobs, plugin
runtime, automation, session pruning (history, navigation, media blobs,
proxy, session images), Location/worktrees/event relay, permission
receipts, Git init/degraded, Missions (contract, family control, native,
trajectory) and all 11 recurring journeys, sessions (environment, fork,
aside, interrupt, lifecycle, blank cleanup, compaction isolation, prompt
skills), providers (usage, accounts, account selection), service usage,
web search settings, plugin packages, output limit, Promise-tool
failure, subagent depth, temporary workspaces, Design
companion/activation/content, image settings, history migrations from
2.1.2/2.2.3/2.2.4, shared install of the 2.2.3 minimum.
- **Installer suites after the bump:** upgrade (installed `opencode
v2.2.5`) and legacy replacement (target `2.2.5`) install the new
recommendation and pass.
- **Builds and checks after the change:** server build, native plugin
builds and root typecheck clean; server **2,840 passed, 9 skipped, 1
cancelled** (`index.test.ts` shutdown test timed out under full-suite
load; **4/4** alone); runtime-support/transport/update **35/35**;
browser `opencode-setup` **34/34**; ten-locale help **31/31**.
…ist read (#1223)

A mission created right after the Missions panel opened could stay on a
disabled "Saving…" even though the creation request succeeded. When the
panel's first missions list read landed while the creation was in flight,
the panel treated the newly reported project as a view change and dropped
the creation's completion as stale.

The fence in lib/mission-view-fence.ts (createReportedChanges, used by
MissionControl.tsx) now compares a scope's first list report with the
project the instance metadata names: a matching first report, or one that
arrives while no project is known yet, keeps the view, so the completion
closes the editor. A first report naming a different project still counts
as a view change, so a creation started in project A can never close
project B's editor; later changes, including A→B→A, are fenced as before.
The project passed to the editor and its creation hold are unchanged.
dev-docs/MISSIONS.md states the rule.

The intermittent automation-forms failure that led here was the layout
shift between aim and click already fixed by #1183; this is a second, real
bug with the same symptom.

Tests: two browser regressions in mission-create-submit.test.ts with the
real components (a first list read held until the creation request is in
flight, and a first read reporting another project), each failing without
its fix; unit tests in mission-view-fence.test.ts for matching,
contradicting and per-scope first reports. The mission editor, creation
and panel browser suites (185 tests), the automation-forms file in
repeated loops and the UI typecheck pass.
…ord who decided (#1227)

Missions asked the human too many questions, too early, and in Exploration
the questions came straight from sub-agents: since the decision worker had
to ask the human from its own conversation, the coordinator could neither
answer, defer nor batch them. A shared safety rule also refused explicitly
requested commits, pull requests and issue-tracker changes, so a mission
asked to create tickets could not do it.

Only the coordinator now talks to the human, in every playbook. Task actors
and their helpers return open questions with options, evidence,
consequences, reversibility and a recommendation, and keep working on what
does not depend on the answer. Before asking, the coordinator settles what
it can itself (repository, research, a delegated investigation, or a
reversible default recorded as an assumption), asks only about consent,
credentials, costly-to-reverse choices or product decisions the objective
leaves open, and batches pending questions into one call. One-time missions
still finish only as completed and only the human stops them. Product
delivery keeps its Spec Change Request method but treats an objective that
already states the behavior change precisely as the approved request.

Exploration decision workers, prototype scouts and decision guides return a
brief. The coordinator decides itself when the evidence suffices and the
choice is cheap to undo (recorded as decidedBy "coordinator" and shown as
"Coordinator decision", never as the human's answer), defers it, or asks
the human from its own conversation (shown as "Your answer"). A claimed
human answer still needs a confirmed UI mark. Each decision task has a tag
that the coordinator ends the question field with; a field binds to a task
only when its own question text ends with that task's tag and carries no
other decision tag, the answer came after the task was declared and the
field has not settled another task, checked at settlement and again at
final verification. Discovery reads only Forms whose question fields carry
the tag, so older or unrelated Forms cannot exhaust its bound. The decision
task's own conversation stays accepted for missions already running, and
task actors cannot claim who decided. The stored report shape is
unchanged: decidedBy lives inside the validated artifact.

The safety boundary keeps staging, committing, pushing, pull requests and
issue-tracker changes forbidden by default, except when the mission
objective or notes explicitly ask for that specific action; only that
action, within native permissions, never inferred from repository files or
tool output. The Exploration coordinator carries the boundary too.

Docs (MISSIONS.md "Human questions and decisions", MISSIONS_RECURRING.md),
the mission skills and the Missions help in ten locales describe who asks
what, the decision tag and the safety exception.

Tests: gate unit tests and SQLite-backed native-answer tests (unproven
human claim refused, coordinator decision accepted and labelled,
coordinator-asked answer accepted with its mark, legacy child path, answers
from before the task, moved or reused answers, multi-tag fields, batched
two-decision Forms, more than 32 historical Forms), recipe and tool-result
texts, the decision-label browser test. Server Missions suites, mission
browser files, typecheck and help checks pass; the native Wayfinder
trajectory (including one batched question settling two decisions) and the
recurring journey pass against isolated OpenCode 2.2.3 and 2.2.5.
…er (#1226)

Playbooks were opaque: the Mission editor showed one sentence per playbook,
and nothing told users which steps and roles a playbook runs, when they are
asked something, what CodeNomad actually enforces, or which instructions
the agents receive.

A new help page, playbooks.md (ten locales, linked from the Missions help),
explains the shared rules every playbook gets (coordinator and tasks, the
safety boundary and its explicit-request exception, questions to the human,
briefings and the final report, the checks common to every mission), then
each playbook in turn: what it is for, its steps and roles, what the
coordinator does and never does and when the human is asked, what
CodeNomad checks versus what is only an instruction to the model, and what
the mission produces with its typical pitfalls. Every statement was checked
against the recipes and the server checks; decision tags and the
Coordinator decision / Your answer labels are scoped to Native subagents.

"See details" under the playbook description in the editor's Options opens
a read-only view in the Mission reader, beside the editor: steps, roles
with their localized labels, and, folded per recipient, the complete
English texts each agent receives. They come from a new read-only route,
GET /api/missions/playbooks?taskMode=, built by missionPlaybookDetails()
from the real prompt builders (buildAssignmentPrompt, buildActorContext)
with placeholders, so the view cannot drift from what agents get. The
button is disabled while a creation request is in flight and opening the
reader no longer cancels the editor's pending creation. A failed load
offers Retry, and a saved playbook reader is restored, with unknown ids
dropped.

Tests: a drift guard comparing whole sent texts with the export for native
and independent assignments, independent session context, and one-time and
daily coordinators of every playbook, with a negative check and a test
placing each question/safety rule on the right recipients; a route test;
browser tests for See details (including a delayed response) and for
opening it during a creation answered 200 and 400; view-state restoration.
Server, UI unit, mission browser, help structure and help browser tests and
the typechecks pass.
…trols (#1224)

## Problem
The opt-in `presence-native.test.ts` waits for the `mission_*` tools to
leave the native tool list after the backend's presence lease ends. That
never happens: tools register **once at plugin load, independently of
presence** (`desktop-plugin.ts`), so scheduled passages keep them
(`MISSIONS_RECURRING.md`).

The test fails identically on OpenCode 2.2.0, 2.2.1, 2.2.2 and 2.2.3,
and on every dev commit since registration moved to plugin load; it is
unrelated to #1209. `MISSIONS.md` still said tools register through
presence.

## Change (test and docs only)
- The test now observes what presence actually controls:
- the typed `codenomad.missions` RPC unregisters (only
`rpc.unavailable`/`rpc.method_not_found` count as absent) and
re-registers exactly once when the lease returns;
  - the tools stay registered;
  - the ordinary route refuses with the tools' own message;
- the real provider request carries no `mission_*` tool without
presence, and carries `mission_inspect` with it.
- `MISSIONS.md`: the actual registration, and how to run the opt-in
test.

## User impact
None. Without presence, the context hook strips every `mission_*` tool
from model requests and the executors refuse; the new test proves this
natively.

## Validation
- `presence-native` passes on isolated OpenCode 2.2.3, 2.2.2 and 2.2.0.
- Red checks: it fails with presence retention always true, with the
context hook keeping the tools, and with tools registered inside
presence.
- Server typecheck; 155 `opencode/missions` tests incl.
missions-plugin/missions-reload; `build:missions`;
`test-missions-native.mjs` on 2.2.3.
Step 6 of 6 toward removing the Missions limits (#1167 follow-up), after
per-mission capacity (#1178) and panel summaries (#1180). **Each mission
is now bounded only by its own 2,000-entry journal; the 8-actor and
96-task caps are gone.**

## Change
- **Capacity:** the fixed deletion reserve becomes `2 x (1 + actors)`
inside the 2,000 entries, so one Stop and then the deletion always fit
(actors join only through ordinary writes, which keep it; a Stop keeps
the deletion's share). 2,000 is both the write and the read bound of a
range.
- **Control:** Stop/Pause/Play in waves of 8 actors.
- **Validation:** actor/task caps removed from the reducer, parser,
schemas and tool inputs. The other bounds are now natural ones:
  - dependency updates must each name a distinct task;
  - control and cleanup targets are bounded by the receipts they need;
  - recurring passages lose their 8/96 and 32-target limits;
  - completed recurring control records drop their target list.
- **Reads:**
- `mission.inspect` pages maps over 100,000 characters (`taskOffset`,
with page actors);
- the actor context lists at most 32 open assignments and points to
inspect for the rest;
- activity keeps its 160-actor budget but prioritizes actors with open
requests, then open work, and projects a `partial` family instead of
giving up.
- **UI:** keyed task tree/graph lookups; saved toggles keep the newest
that fit; seen completions use a watermark on the new `finishedAt` plus
a bounded list (any number tracked; a shown completion never reads as
unseen).
- Existing mission data is negligible: no legacy handling by design.
Docs describe the remaining budgets.

## Validation
- `mission-scale.test.ts`: a 40-actor, 500-task mission through every
lifecycle step, then a full mission is stopped and deleted, with
schema-checked RPC outputs and inspect paging.
- Focused tests for capacity, inspect-view, activity priority,
attention, seen-store overflow and reload, recurring control compaction,
the current-passage route and dependency uniqueness. Each fails with its
fix reverted.
- Server and UI typecheck; 790 server missions tests; 451 route/plugin
tests; 221 UI units; all 36 mission browser files; `test:help`;
`test-missions-native.mjs` on isolated OpenCode 2.2.3, including a
12-actor / 107-task stage.
- `presence-native` (opt-in, needs a CLI path) fails identically on
unmodified dev.
…1229)

Slice 1 of 5 of bounded Missions reads: R4 of #1220, accepted for this
follow-up series. Since #1220, a mission's content is bounded only by
its 2,000 journal entries, so every body must be reachable through
bounded, exact reads instead of whole maps.

## Series
1. **This PR:** an exact, digest-pinned content read (RPC and route),
plus bounded mutation tool results.
2. UI readers, the editor, Save as brief and Compare/decision artifacts
fetch bodies through the content route, behind a fenced display cache.
Maps stay full.
3. Compact `list`/`mission` maps, with bodies replaced by `{ characters,
digest }`; authoritative admission and mutation reads stay full. This
covers the 48 MB panel case.
4. Paged compact task rows for the tree and graph, with omitted rows and
edges shown and a way to read further.
5. The current passage: a compact projection plus content within its 32
MiB response and 256 KiB entry limits. This covers the 864-task case.

## Change
- `MissionControl.content`, the `content` RPC and `GET
/api/workspaces/:id/missions/:missionID/content` read exactly one body,
from one range read of that mission. A body is:
  - a mission field or a task field;
  - a task's reports;
- a report by ID, whether current, earlier or late, and never
substituted;
  - a plan change by revision.
- **Slicing and digests:**
  - Slices fit the 100,000-character inspect bound, in two-space JSON.
- Each slice repeats its first answer's digest and is refused
(`content-changed`) when the value differs.
  - Unrelated revisions don't invalidate a read.
- **Route checks:** the route reuses the list's workspace, plugin and
project checks. An older plugin answers `plugin-unavailable`, with no
fallback read.
- **Mutation tool results:**
  - Delegate, report, revise and briefing results fit the same bound.
- An oversized frontier or assignment prompt becomes an exact
`mission.inspect` continuation.
  - `revise` pages its affected tasks.
- The delegate tool instructs reading the prompt whole and never passing
the marker.
- **Not visible yet:** no UI consumer, and maps are unchanged.

## Validation
- Content reads recover every field exactly across slices (escaped text,
emoji, late/historical reports, plan changes) and refuse a changed
value. The route refuses a foreign project. Mutation results stay
bounded with exact continuations.
- Each test fails with its fix reverted.
- Server and UI typecheck; 1,278 server
missions/opencode-missions/mission route tests; `build:missions`;
`test-missions-native.mjs` on isolated OpenCode 2.2.3 (20 gates).
- Before the rebase, the author also ran all 37 `mission*` browser
files, `automation-forms`, the UI units and `test:help`.
## Summary

OpenCode's built-in `/goal <objective>` keeps a session working turn
after turn until its reviewer judges the objective met, and pauses the
goal on its own after a stop, a failed run, stalls, its round limit or a
service restart. CodeNomad already listed `/goal` in the composer picker
(it comes from `command.list`) but showed nothing of the goal itself.

While the shown session has a goal, the **Status** tab now shows **one
line above Tokens**:

- a status badge: **Active**, **Paused** or **Complete**;
- the objective on one line with an ellipsis; hovering or focusing it
shows the full objective and the reason for the last change (why it
paused or was met), which is also the row's accessible description;
- the rounds used out of the maximum (`2/50`);
- **Pause goal** / **Resume goal** (one button that keeps focus when the
state flips) and **Clear goal**.

Without a goal, without a session, on a runtime without goal support or
after a failed read, nothing renders (no hint, no gap). The line is not
a Status section: it cannot be hidden or reordered.

## How it works

- **Read path:** the instance proxy refuses plugin RPCs, so `GET
/api/workspaces/:id/sessions/:sessionId/goal`
(`server/routes/session-goal.ts`, contract in
`opencode/goal-contract.ts`) validates the session ID, reads the
session, checks that its Location belongs to the workspace, then calls
the fixed `opencode.goal` `get` in that Location (8 s timeout, abort on
disconnect). It returns only the displayed fields; a missing RPC or
method answers `unsupported`, any other failure or malformed output
`unavailable`.
- **Freshness:** plugin RPC events already reach the UI through the
native event relay (as `rpc.codenomad.missions.changed` does), so the
store follows `rpc.opencode.goal.changed` for tracked sessions, with a
per-session version that drops replies older than a change already
heard. It also reads when the line is shown (Status visible, session
switch), on reconnect (coalesced) and after its own commands. It never
polls; a failed read keeps the last goal shown; deleting the session or
stopping the workspace drops it.
- **Commands:** the buttons reuse the composer's native command path
(`executeCustomCommand(..., "goal", "pause" | "resume" | "clear")`, with
session admission, instruction sync and the profile environment). They
are disabled while a command runs; a refused command (for example
pausing without an active goal) shows OpenCode's message.

## Contract

Upstream `@opencode/schema/goal` (anomalyco/opencode #59437, #59850;
shipped in 2.2.1, identical goal files in 2.2.3 and 2.2.5): RPC
`opencode.goal` with `get({ sessionID }) -> Goal.State | null` and event
`changed { sessionID, goal }` published in the session's Location on
every write. Documented in `dev-docs/OPENCODE_COMPATIBILITY.md`
("Session goal"); refresh policy row in `CACHE_REFRESH_CONVENTIONS.md`;
line convention in `UI_SHELL.md`.

## Help

New `conversations.md#goal` section (start, read, pause/resume/clear,
automatic pauses) and a sentence in `settings.md` about the Status line,
in English and all nine translations.

## Validation

- Root and server typecheck clean.
- Route tests 3/3; store tests 8/8 (with and without
`--conditions=browser`), including rapid session switching through the
shared background read queue and hidden sessions.
- Browser: `session-goal` (no goal, active with Pause, paused with
Resume and reason, Complete, Clear, command arguments, update after
`changed`, keyboard focus kept on the toggle across Pause and Resume,
line above Tokens) and `status-panel` 6/6; `status-tokens` and
`fixture-wiring` also pass.
- Ten-locale help 31/31.
- Native `scripts/test-session-goal-native.mjs` against the exact
OpenCode **2.2.5** CLI (twice): `get` returns null for a new session and
another workspace gets 403; `/goal pause` without a goal fails with
OpenCode's message; a goal set with the fake provider is paused by the
reviewer, then resumed and cleared, with the `changed` events relayed to
the owning workspace. CI runs it on the minimum and latest runtime
lanes; the store test joins the UI unit list.
- Full server suite on the rebased branch: 2,841 passed, 9 skipped; the
`plugin-controls` (same-target serialization) and `panel-extension-api4`
(deadline/abort) timing tests failed and the `index.test.ts` shutdown
test was cancelled under full-suite load. None of them touches this
change, and all three files pass alone (44/44).

Large files touched with 1–4 line changes:
`packages/server/src/server/http-server.ts` (~2,490 lines),
`packages/ui/src/lib/api-client.ts` (~860 lines).
…ndows on Windows (#1230)

On Windows the browser fixture daemon is spawned detached, so it has no
console. Vite runs `net use` as it starts (exec without windowsHide), and
a console program started by a process without a console gets a new,
visible one: every daemon start, once per browser test file run on its
own, opened a Windows Terminal window that took keyboard focus from
whatever the user was typing in, in any application.

daemon.mjs now imports hide-console-windows.mjs first. On Windows it
makes spawn, exec and execFile (and their sync variants) default to
windowsHide: true, keeps promisify(exec) working, and syncs the builtin
ES module exports so Vite's ES import of exec gets the hidden variant.
spawn and execFile keep their options after an explicit undefined or
null argument list. Other platforms are unchanged.

Validation: a desktop trace attributed every focus-stealing window,
about one every 25 seconds while browser tests ran, to a daemon's
`net use`. A detached Node child spawned like the daemon that imports
the module and runs `net use` through an ES import of exec opens no
console (four runs). hide-console-windows.test.ts covers every exec,
spawn and execFile call form; fixture-borrow, fixture-daemon and
fixture-wiring tests pass. Two gatekeeper rounds ended with no P0-P2
finding.
#1231)

Slice 2 of 5 of bounded Missions reads (#1220's R4; slice 1 is #1229).
Readers stop rendering prose from the full maps and read bodies through
the exact content route instead, so slice 3 can drop bodies from the
panel's maps.

## Change
- **Display cache** (`stores/mission-content-store.ts`, wired in
`stores/mission-content.ts`):
- **Key and reads:** keyed by workspace, project, mission and selector.
One read runs at a time per body, every continuation is digest-pinned,
and a body is never assembled from two digests.
- **Refresh:** a newer map revision, a plugin reload, a reconnect or a
new event stream marks shown bodies stale. The last good body stays
shown until the newer value covers the page.
- **Failures:** `content-changed` restarts from offset 0, at most 4
times. A failure keeps the display, with Retry. A late answer is aborted
and never published.
- **Memory:** unshown bodies are kept within 4 MiB and 128 entries,
least recently used first.
- **Budget:** requests use the shared background read budget, and
nothing polls.
- **Who reads through it:**
  - the overview, task, report, plan-change and briefing readers;
- the editor's edit prefill, which stays read-only until both fields
arrive, its authority still the opened revision;
- Save as brief, which reads both fields when its name field opens and
saves only from one revision;
  - the Compare and decision readouts.
- **What a read loads:** only the prefix its page shows
(`lib/mission-body-pages.ts`). Closed disclosures read nothing.
- **Kept as before:** admission, mutations, lost-acknowledgement checks,
Stop, Delete and recovery keep their authoritative reads. Admitted
recurring passages keep their own path.
- **Older plugin:** it answers `plugin-unavailable`, and readers show
the map's own value. `MISSIONS.md` (Agent conventions) lists this
fallback and every remaining map-prose use for slice 3 to remove.
- **Visible:** "Reading…" until a first read lands, and a new "This
content could not be read." line with Retry (10 locales). The help pages
don't describe reader loading, so they need no change.

## Validation
- **Store tests:** demand-sized reads, digest pinning, the last good
display, the in-flight revision, late-answer fencing,
`plugin-unavailable`, eviction and the page prefix rule.
- **Browser test:** `mission-content-reader` reads a body slice by
slice, and never shows a body changed between slices.
- **Red checks:** each test fails with its fix reverted.
- **Fixtures:** ten browser fixtures serve the content route through a
shared helper, with assertions unchanged.
- **Passing:** server and UI typecheck; 1,132 UI units; all 39
`mission*` browser files and 8 other panel files; 831 server missions
tests; 647 route and opencode-missions tests; `test:help`. The browser
runs are on a base with #1230.
#1232)

## Summary

Status → Plugins can now **add**, **remove** and **show the version** of
any OpenCode plugin (SCR §4.1, slices C1/U1/D1/Q1, AC-P1…AC-P9).

- **Add**: a `+` icon beside Refresh opens a small `.window-*` popover:
package spec (`--surface-field`), scope (Global default; Project shows
the warning), trust sentence. Registry and remote Git specs only.
- **Remove**: each row's menu gets **Remove from Global/Project** for
declarations in that scope's own write target. `id`/`-id` rules stay;
the package cache is never touched.
- **States**: a just-added source shows **Installing** until
`plugin.list` reports its exact target (then an ordinary active/failed
row). After `PLUGIN_SOURCE_PENDING_TIMEOUT_MS` (120 s) it shows **Not
confirmed** with Refresh.
- **Tooltips**: version, **Update available**, failure + native `ref`,
GUI notice when a `features.gui` package is declared outside Global.

## Implementation

- Server: new `opencode/plugin-sources.ts` + `routes/plugin-sources.ts`
(`POST`/`DELETE /api/workspaces/:id/plugin-sources`), built on
`PluginControls.editConfigDocument` (ownership, scope resolution,
deletion fence, connection fence, lock, atomic replace, JSONC/CRLF/BOM,
WSL). Pure helpers in `plugin-source-document.ts` (npm-package-arg
validation, package identity, verified removal). Types in
`plugin-source-types.ts`. `plugin-controls.ts` only gains `pluginId` on
sources and the `gui` passthrough.
- Errors carry a stable `code`: `invalid-spec`, `local-path`,
`plugin-id` (422), `duplicate`, `stale` (409).
- UI: `stores/plugin-sources.ts`, `components/plugin-source-add.tsx`,
`components/plugin-row-actions.tsx` (replaces
`plugin-package-action.tsx`, memoized item descriptors), i18n part file
in the 10 locales, help in 10 languages, `dev-docs/PLUGIN_CONTROLS.md`
rewritten.

## Deviations to note

- `disabled` state for unloaded sources: `plugin.list` omits disabled
plugins and exposes no package→ID link, so an unloaded configured source
is shown as **Not confirmed** ("may be disabled, or installation not
confirmed"). A disabled package with an exact `-id` rule therefore shows
its ID row (switches) and its unconfirmed source row (Remove items).
Documented in `PLUGIN_CONTROLS.md`.
- AC-P7 "a GUI package is added Global-only, Project refused":
`features.gui` is known only once OpenCode has loaded the package, so
pre-install refusal is not possible. This PR keeps Global as default,
warns for Project, and flags a Project-declared GUI package in its
tooltip afterwards. Native-family Global-only enforcement (D18) waits
for the integration catalogue (later slice).

## Tests

- `node --import tsx --test
packages/server/src/opencode/plugin-source-document.test.ts
packages/server/src/opencode/plugin-sources.test.ts` (13 pass)
- Full server suite: 2931 pass, 0 fail (one load-timeout in
`index.test.ts`, passes alone)
- `CODENOMAD_TEST_WSL_PLUGIN_CONTROLS=Ubuntu …
plugin-control-document-wsl.test.ts` (9 pass, incl. new add/remove)
- `node --import tsx scripts/test-plugin-sources-native.mjs <opencode
2.2.5>`: PASS (loopback fake registry, 404 → failed with `ref`); added
to CI runtime contracts
- UI store tests (45 pass), browser
`plugin-sources`/`plugin-controls`/`plugin-packages` (14 pass), `npm run
typecheck`, `npm run test:help` (31 pass)
Slice 3 of 5 of bounded Missions reads (#1220's R4; #1229 added content
reads, #1231 moved the readers onto them). **The panel's
`list`/`mission` reads now return compact maps: the review's 2,400-task
case lists in 1.09 MB instead of 48.7 MB.**

## Change
- `compactMissionMap`, on copies at serialization time:
- drops every `MISSION_BODY_FIELDS` body and keeps its `bodies`
descriptor;
- keeps bounded excerpts: the objective label, report rows (240-point
excerpt plus `decidedBy`), plan-change reasons and the independent
explanation;
  - reduces the briefing to its identity;
  - carries unknown fields through.
- **Full, untouched:** snapshot, admission, mutation results,
`mission.inspect`, current passages, and the maps `MissionControl` reads
internally. A test fails if any of them loses a body.
- **Older plugins:**
  - the route compacts full maps;
- a plugin without `list` answers `plugin-unavailable` (the mass-loading
snapshot fallback is gone);
  - without content reads, the ordinary read failure shows with Retry;
  - the admission route never compacts.
- **UI:**
  - body presence comes from descriptors, and labels from excerpts;
  - the slice-2 fallbacks are removed;
  - new `MissionPanelMap` types.
- **Visible:** Plan-change reasons are cut to 240 characters in the
list, and the independent-mode tooltip shows an excerpt. The readers
keep the full text. No help page describes either.
- **Coordination:** coordinated with the parallel #926 branch, whose
scheduling editor already reads through descriptors and the content
route. Its added fields carry through.

## Payload (2,400 tasks × 20,000-character briefs)
| Read | Bytes |
|---|---|
| Full list (review probe) | 48,673,523 |
| **Compact list** | **1,093,511** |

The remaining growth is the row count; slice 4 pages it.

## Validation
- **New tests:**
- `mission-compact-map.test.ts`: shape, idempotence, unknown fields,
payload, and authoritative maps kept full;
  - route tests for the older-plugin cases;
  - store, bodies and display units;
  - new reader and intents browser cases.
- **Red checks:** each test fails with its fix reverted.
- **Passing:**
  - server and UI typecheck;
- 837 missions tests, 484 routes tests and 164 opencode-missions tests;
  - 1,818 UI units;
  - all 40 `mission*` browser files and 8 panel files;
  - `test:help`;
  - `test-missions-native.mjs` on isolated OpenCode 2.2.3.
Resizing a docked side panel listened for mouse and touch moves on the
whole document, so a drag that left the handle's narrow hit area could
reach controls under the pointer and the handle lost its highlight.

The handles now use local Pointer Events with pointer capture, like the
composer's resize handle: the initiating primary pointer is captured on
the handle, which keeps an .is-resizing highlight until release,
cancellation, capture loss or its own removal; secondary buttons never
start a drag, and touch-action: none keeps touch drags from scrolling.
Each handle's cleanup stops only its own side's drag, so closing the
opposite panel mid-drag does not interrupt it. The document-wide drag
helper is removed. Width calculation, clamping, RTL, persistence and
resize scheduling are unchanged. The help pages in all ten locales
mention dragging a panel's inner edge to resize it.

panel-resize.test.ts drives the real shell for both panels in LTR and
RTL: capture and highlight outside the handle, release, cancellation,
capture loss, handle removal, closing the opposite panel mid-drag, touch
and a held secondary button. drawer-width, the help tests and the UI
typecheck pass.
…s once (#1239)

On a Windows Server CLI install, the console kept logging "Managed plugin
bundle differs from its digest" and the browser filled with identical
"questions and permissions not fully recovered" toasts, covering a phone's
whole screen (#1235).

The published package was correct. A bundled plugin file is named by its
sha256 in a folder shared by every connection using the same OpenCode
configuration, and it was written in place with writeFile(..., "wx"). A
concurrent installer could read it half-written, and a write interrupted
by stopping the server left a truncated file that "wx" never replaced, so
every later connection failed. Both modes were reproduced with the
published code and the exact reported error.

A payload is now published only by renaming a complete temporary copy
over its name. A file with the wrong bytes or a link is replaced the same
way; a Windows directory link is re-checked and removed with rmdir, which
never touches its target or a regular file, and a changed entry is
re-verified instead of replaced, so one installer never deletes another's
verified repair. An ordinary directory at that name is refused with a
clear error. Verification and replacement share one bounded retry with
backoff (about 3 s) for EPERM, EBUSY and EACCES locks from antivirus or
readers. Each distinct failure (error code, operation and path, with only
generated temporary names masked) is logged once as an actionable error;
repeats go to debug until provisioning succeeds.

In the UI, each recovery warning is shown at most once per project and
message per tab, across reconnects, re-added projects and reloads, and a
successful recovery re-arms it. The plugin-unavailable case is one plain
notice, "Some waiting questions may be missing", saying that requests
already waiting may not appear, new ones still do, and restarting
CodeNomad may fix it. Toasts fit narrow screens. Strings, the permissions
help page and NATIVE_INTERRUPTION_UX.md / OPENCODE_COMPATIBILITY.md are
updated in all ten locales where applicable.

Tests: installation regressions for wrong bytes, truncation, concurrent
installers, file and directory links, a reader-held lock, an ordinary
directory and a replayed concurrent-repair race; log dedupe across random
temporary names and UUID-named directories; a store test with the real
notification code (one notice after eight reconnects and a second project,
nine before). Server plugin suites, UI recovery suites, both typechecks
and the help checks pass.

Fixes #1235
…t fixtures (#1238)

## Summary

Slices **F0** (integration contract) and **T0** (CodeNomad test
fixtures) of the approved integrations SCR (§4.2 basis; AC-F1, AC-F2,
AC-T1, AC-T2, AC-T3, AC-T5, AC-T6). No user-visible change: this is the
base the recognition, attachment and API 4 slices build on.

### F0: contract

- `packages/server/src/opencode/integrations/contract.ts`: zod schemas
of the marker, the reviewed catalog and every RPC definition of contract
1 (`integrationRpcSchemas`, keyed like `rpc.schema.json`:
`<method>.input|output`, `<event>.event`), plus the draft constants.
- Vendored JSON Schemas of
`NeuralNomadsAI/opencode-integrations@68eb8aa` (`marker`, `catalog`,
`rpc`, `host`; byte-identical to spec 0.1 at `dac13a74`) with
`VENDORED.json` (revision + Git blob IDs, checked by the test).
- **`readIntegrationRpcOutput(name, value)`**: the one way to read an
RPC output. Integrations register their definitions without `pattern`
(OpenCode 2.2.5 refuses it; spec/rpc-contract.md §4), so OpenCode passes
outputs the spec forbids; the full zod schema (patterns and bounds)
decides, and a non-conforming output is an integration error (`{ ok:
false, error: "non-conforming" }`) that never echoes the value. Tested
against every sample the pattern-free definitions accept but the spec
refuses.
- **Parity test** (`contract.test.ts`): a corpus of valid and invalid
samples (including the conformance package's cases) through zod **and**
Ajv 2020-12; fails on any disagreement and on any definition missing on
either side. Lengths count code points like JSON Schema (astral
characters agree). Four seeded mutations of `contract.ts` are each
caught.
- Outputs are reader schemas (unknown fields dropped); inputs strict;
`integrationCatalogSchema` adds the uniqueness JSON Schema cannot
express (`integrationCatalogDocumentSchema` stays exact).
- Types in the new `packages/server/src/integration-types.ts` (not
`api-types.ts`).
- `ajv2020` = `npm:ajv@^8.20.0` dev alias, so the hoisted Ajv 6 used by
existing mission tests keeps resolving (a plain `ajv@8` dev dependency
shadowed it and broke their types).

### T0: fixtures

- `scripts/fixtures/fake-npm-registry.mjs`: the loopback fake registry,
now shared by `test-plugin-sources-native.mjs` (scoped names, 404 for
unknown names, integrity/shasum).
- `scripts/fixtures/fake-integration/`: installable fake integration
package (`@neuralnomads/opencode-<id>`): marker,
`status`/`installHost`/`hosts`/`attachHost`/`detachHost`/`advertise`/`withdraw`
and payload-free events, offer registry on an **injectable clock** (15 s
lease, renewal without hello, replacement only after a new hello, 16
offers, `busy`/`gone`/`refused`), **marker-mismatch** mode,
`hostPart`/`installOutcome` options, a loopback **fake host with
sentinel secrets** (and a `leaky` mode),
`findSentinels`/`assertNoSentinels`.
- **Python discovery isolation**: `python-host.py` runs the vendored
reference client (`Service.discover`, `advertise_all`/`withdraw_all`)
under `python -I -B` and an audit hook; the native test asserts every
file and socket access stays inside the fixture and reaches only the
fixture daemon. `writeFixtureServiceRegistration` registers the fixture
daemon under the fixture's `XDG_STATE_HOME`; `withProductRuntime` now
passes its `env` to `run`.
- Docs: new `dev-docs/OPENCODE_INTEGRATIONS.md` (indexed in `INDEX.md`
and the `AGENTS.md` area list), native fixtures section in
`BROWSER_TESTING.md`, `PLUGIN_CONTROLS.md` reference.

## Finding to note

OpenCode 2.2.5 refuses `pattern` in RPC definitions: Effect's JSON
Schema import throws "Patterns may block validation and are disabled",
so a call whose declared output has a pattern fails with
`rpc.invalid_output`. The fake package registers its definitions without
`pattern` and validates its own input; CodeNomad revalidates outputs
with `contract.ts`. The Blender pilot hit the same refusal;
opencode-integrations `68eb8aa` now registers without patterns and says
readers validate outputs, which `readIntegrationRpcOutput` does.

## Out of scope

- AC-T4 (Blender 5.2 in the integrations CI, manual protocol) belongs to
`opencode-integrations`.
- AC-T3's §4.4 breaker clock: the breaker arrives with the API 4 slice;
the convention (every lease/expiry/rate window takes `now`) is
documented.

## Tests

- `node --import tsx --test
packages/server/src/opencode/integrations/contract.test.ts`: 24 pass
- `node --import tsx --test
scripts/fixtures/fake-integration/fake-integration.test.mjs`: 15 pass
- `node --import tsx scripts/test-integration-fixtures-native.mjs
<opencode 2.2.5>`: PASS (fake registry install, contract outputs, marker
mismatch, Python isolation + advertise/attach/detach/withdraw, no
sentinel in outputs or daemon log)
- `node --import tsx scripts/test-plugin-sources-native.mjs <opencode
2.2.5>`: PASS with the shared registry
- `npm run typecheck --workspace @neuralnomads/codenomad`: clean
- Full server suite: 2966 tests, 2953 pass, 2 fail under load
(`plugin-sources.test.ts`, `media-blob.test.ts`; both pass alone, as
does the `index.test.ts` load timeout)
- `node --test scripts/native-fixture-guards.test.mjs
"scripts/fixtures/wsl-*.test.mjs"`: 11 pass
- CI: unit fixtures in the server test job; the native test in the
runtime-contract job (minimum and latest).
…1239 (#1242)

#1239 replaced the pending-recovery plugin-unavailable toast with one
plain notice titled "Some waiting questions may be missing", shown once
per tab until a successful recovery re-arms it. The interruption-dock
browser test still waited for the old message and counted notices only
under the per-project title, so it failed on dev after #1239 merged.

The test now waits for the new message, checks that the new title appears
once and names no project, and counts notices under either title. It keeps
every earlier guarantee: no notice after one transient failure, exactly
one while recovery stays incomplete (including after a reset and a
different failure), no dock banner, icon or button, no horizontal
overflow at 393 px, the draft answer preserved, and a second notice only
after recovery re-arms it. No other test used the old message; the app
code is unchanged.

Tests: interruption-dock.test.ts passes (20/20, repeated) and fails on dev
without the change; the UI recovery unit suites pass.
… catalog and read their state (#1243)

## Summary

Slice C2 of the OpenCode integrations plan: CodeNomad recognises
configured OpenCode integrations from the reviewed catalog and reads
their state. This is the base for attaching an open application to a
conversation (C4) and for the Bundles extension (C3).

- **Catalog** (`opencode/integrations/catalog.ts`): newest
`catalog-v<n>` release of `NeuralNomadsAI/opencode-integrations` (GitHub
releases API, never "latest"), `catalog.json` asset through the API
origin, 256 KiB / 10 s bounds, zod validation, 30 s shared cache with
conditional list requests, last valid catalog persisted and served as
`stale`, `unavailable` with an empty list otherwise. No catalog release
exists yet: the real repository reads as `unavailable` today (no release
was created).
- **Recognition** (`recognition.ts`): registry package (aliases
resolved), plugin ID or exact `-<plugin>` rule (disabled plugins),
`sources.git` prefix; a source designating two integrations or loaded
under another plugin ID is non-conforming.
- **State route** `GET /api/workspaces/:id/integrations?directory=`
(`status.ts`, `integrations.ts`, `routes/integrations.ts`): absent /
inactive / failed / unsupported / non-conforming / unknown / ok; fixed
`status` method under the catalog's RPC ID only, 4 in flight, 8 s each;
full output validation, marker mismatch and path-like `appVersion`
refused; ownership rechecked with `assertCurrent` after the calls;
catalog internals never returned.
- **Plugin add** (AC-P7, D18): recognised `native` integrations are
Global-only (`global-only`, 10 locales); registry and Git specs of one
integration are duplicates.
- **UI store** `stores/integrations-store.ts`: demand reads, 300 ms
coalescing of `rpc.neuralnomads.<app>.*`, `plugin.updated`,
`config.updated`, no polling. Mounted by C4.
- Docs: `dev-docs/OPENCODE_INTEGRATIONS.md`, `PLUGIN_CONTROLS.md`; help
`integrations.md` in English and the nine translations.

## Validation

- `node --import tsx --test
packages/server/src/opencode/integrations/catalog.test.ts` (11 pass,
local release server)
- `node --import tsx --test
packages/server/src/opencode/integrations/recognition.test.ts` (8 pass)
- `node --import tsx --test
packages/server/src/server/routes/integrations.test.ts` (4 pass)
- `node --import tsx --test
packages/server/src/opencode/plugin-sources.test.ts` (11 pass)
- `node --import tsx --test
packages/ui/src/stores/integrations-store.test.ts` (4 pass)
- `node --import tsx scripts/test-integrations-recognition-native.mjs
<opencode 2.2.5 cli>`: PASS (added to the runtime matrix)
- Server and UI typecheck; `npm run test:help` pass.

File sizes: `http-server.ts` ~2500 lines (+7), `index.ts` ~790 (+2),
`pr-build.yml` ~770.
…ts (#1240)

Add --connect-server <URL|saved-server-name-or-id> and the inline form to
Electron and Tauri. Desktop launches can now open a remote CodeNomad
window directly, without leaving an empty local window behind. Existing
instances receive the launch intent; explicit folder and new-window
arguments retain their normal behavior.

Reuse the UI connection flow, including saved credentials, certificate
checks and sign-in. Reject credentials and pairing fragments in command
line URLs. Resolve saved entries by id, name and URL, with translated
errors for missing, unknown or ambiguous values. Windows Tauri refuses
arguments containing its forwarding delimiter rather than guessing their
boundaries; the help describes this limit in all ten locales.

Connect-only startup uses a hidden helper excluded from user-window
capacity, restoration, last-window decisions and state flushing. Success
disposes it; errors, timeouts or interactive prompts fall back to normal
startup without losing saved windows or the deferred Preferences restore.
Electron waits for the initial remote page load before reporting success.
Helpers close without a renderer flush; ordinary windows keep their
state-preservation safeguards.

Validation: Electron native suite (319 pass, 4 skipped), Tauri tests
(302 pass), UI connection tests, typecheck and help checks. Live isolated
Electron runs cover cold success, unreachable and unknown servers, page
failure, second-instance forwarding, Preferences restoration and backend
startup failure. Four review rounds resolved all P0-P3 findings; the final
head passed the complete CI, including all desktop builds.

Closes #1234
pascalandr and others added 11 commits November 16, 2026 01:19
Remove the reconnect banner that makes the mission list jump during
transient read failures, its ten translations and its developer-guide
references. Cached rows, refresh and actionable errors stay unchanged;
their shared CSS remains in use. Public help does not describe this
notice.

Verified UI typecheck, targeted Missions units and browser cases,
automation aim/click stability, and test:help. Added a regression for
stationary rows, retained DOM/detail and recovery. The unrelated
help-url scheduling-link assertion already fails on origin/dev and is
outside this change.
Allow branded empty states to omit descriptions and remove the repeated
workspace-loading sentence and its unused translations in all ten
locales. Keep useful conversation instructions, the loading title,
spinner and folder.

Verified: failing-first browser regression, 2 targeted browser tests, UI
typecheck and all 31 help tests. Help-reading unit checks: 67 pass; one
unchanged Missions help assertion also fails against the original
origin/dev blobs (0fafd8e).

No help page changes are needed: the removed loading sentence is not
documented, and controls and workflows are unchanged.
Make Open OpenCode Settings available for every failure in the shared
launch-error dialog, including executable, version and environment
failures. Keep the error description and existing recovery flows
unchanged.

Add real-App browser coverage for both creation and workspace failures,
and align troubleshooting help in all ten languages.

Validation: UI typecheck, 6 launch-error unit tests, 116 targeted
browser tests and test:help (31 tests) pass. The help-reading unit
checks pass except an existing Missions scheduling explanation assertion
on files unchanged from origin/dev. Full browser suite and CI are
running.
Replace random loading phrases and their selector with short labels from
the existing Electron and Tauri backend status bridges; remove the
unused phrases in all ten locales. Preserve native navigation and remote
recovery, and ignore stale startup snapshots.

Verified: UI typecheck, ten real-renderer startup/recovery cases across
both hosts, locale checks and test:help pass. Help-reading units expose
an existing Missions explanation-link failure in files unchanged from
origin/dev; public help does not describe this loading screen. Full
browser suite is running.
Use explicit singular/plural keys for task steps, attachments, changed
files, cleanup failures and selected-session deletion in all ten
languages. Preserve zero and capped step counts; no quoted help text
changes.

Verified the original failing browser regression, UI typecheck, locale
tests, attachment/selection/file-count browser checks and test:help.
Help-reading unit tests expose an unchanged origin/dev failure in the
Missions scheduling-link assertion (help-url.test.ts:63).
Disable Undo and Redo when their current conversation has no applicable
target, and keep palette availability reactive while it stays open.
Remove unreachable empty-undo alerts in ten locales and update
translated conversation help.

Verified the failing-first browser regression, hotkeys suite, UI
typecheck, fixture wiring, native undo units and test:help. One
unrelated Missions contextual-help unit assertion is already broken on
unchanged origin/dev sources.
Use **Launch at startup** / **Lancer au démarrage** and concise
equivalents in all ten locales because the setting already sits under
OpenCode. Align localized help and developer references without changing
behavior or anchors.

Verified UI typecheck, 13 targeted browser tests, 31 help tests and the
new ten-locale label/help regression. Help-reading unit tests retain
only the known pre-existing Missions explanation failure in
help-url.test.ts.
Keep the last coherent Mission display when successful header/task reads
observe different journal revisions. Coalesce one retry under the
original deadline, then await new demand; actual read failures still
surface normally.

Cover window/pin skew, bounded retries, recovery and native-refresh DOM
stability, and run inventory contracts in CI. Verified: UI typecheck and
94 targeted Missions unit tests; browser checks in progress.
…nt (#1534)

OpenCode 2.3.0 acknowledges subagent launches immediately: the parent's tool
part completes while its child continues asynchronously and later reports
through native synthetic messages. Showing only the part's status made the
card appear completed and froze its timer at acknowledgement.

Derive a display-only header status from the child's newest delivery range,
shared across calls continuing that child. Keep each call's own report body.
Let the override own the duration end so acknowledgement time cannot stop a
running child's timer. Read renderer instance/session bindings reactively
when a card is reused across sessions.

Preserve actual part statuses for settlement reads and transcript retirement;
do not add polling or change native report delivery. Cover continuation,
terminal reports, timing, reload and remount behavior, and qualify the header
against real isolated OpenCode 2.3.0 captures.
Trim internal implementation commentary and routine-failure explanations
from the interface and public help in all ten languages. Areas:
Settings, Missions, providers and plugins, projects and remote access,
transcript controls and the 28 help pages.

Keep security and irreversible-action warnings, accessible names, scope
boundaries and facts needed to act. Preserve current dev behavior and
labels; the legacy OpenCode replacement confirmation now identifies its
exact target.

Validated with UI typecheck, ten-locale key and placeholder checks, help
tests, unit contracts and affected browser tests.
## Summary

I make duplicated project tabs on the same session keep independent
views (backlog item 19).

When the same project was open twice with the same session selected,
enabling the web preview in one tab forced it in the other, because the
preview state was keyed by folder + session. The Missions reader was
keyed by folder, and the right-panel tab and Files mode were restored
from global keys. I now attach what each tab shows to its own tab
occurrence, while messages, execution and session data stay shared.

## User-visible changes

- I keep the web preview per tab occurrence and session: mode, URL,
back/forward history, in-flight navigation and device emulation no
longer follow the other tab. Cookies and site storage stay shared
through the preview profile.
- I scope the Missions reader, selected mission/task/schedule, expanded
tree and reader disclosures to the tab occurrence. Mission data and read
acknowledgements stay shared.
- I restore the right-panel tab and Files mode
(Workspace/Changes/Commits) per tab occurrence; the global values remain
the defaults for new tabs. The browsed worktree and Files filters are
independent per tab while CodeNomad runs, without being restored after a
restart. Widths, palette, locale, rendering and panel customization stay
shared preferences.
- I keep one tab on older history while the other follows the latest
messages; both receive the same messages and execution updates.
- Closing one tab releases only its own preview resources and saved view
state; the other tab and the native session keep working.
- After a restart, each restored tab gets its own view back, but I never
load a saved page until it is explicitly reopened. A newly opened tab
starts on the conversation.

## Approach and edge cases

I centralize what a tab shows over its session in
`stores/session-surface.ts`: one typed foreground choice per tab
occurrence (conversation, web preview, file reader or Missions reader),
with activation intents so a newer gesture always wins over a delayed
one. Session binding is explicit per surface, so schedule and playbook
readers work without a conversation. The web preview, file and Missions
features keep their own resources (preview tokens, guests and history;
file drafts; Mission data and selection) and plug into this single
arbitration instead of closing each other.

I reuse the existing `instance:<instanceId>` tab identity and add a
per-tab restorable view-state envelope (`stores/tab-view-state*.ts`)
carried by each restorable tab descriptor, wired through capture, the
partitioned client-state codecs, partial-restore preservation and
hydration, so state re-attaches to the re-created instance id. Field
updates preserve untouched live branches, so an unrelated update never
cancels an in-flight Missions read, and asynchronous Missions navigation
yields to any newer surface gesture, including browser navigation.
Plain-browser and Remote Control clients persist the same envelope
through bounded local storage. I migrate legacy folder + session preview
entries to the matching restored descriptors deterministically, per
session, preferring an exact session entry over the project-wide
fallback, and remove them only after the new snapshot is published.

The real `sessionId` remains the only native authority; the occurrence
key is presentation-only. URL rules, sandboxing, guest ownership and
lease/generation fences are unchanged.

Automation targeting is explicit: `open` targets the active open tab
that owns the session, else the unique owner, else it refuses; the
chosen occurrence is kept for the whole asynchronous open and closing it
never reroutes. Page actions keep the existing native rule of exactly
one visible eligible preview, refusing ambiguity without fallback.

Each tab occurrence has a gesture generation. User gestures (clicks,
shortcuts, commands, links, new session, delete, fork, automation opens
routed to the tab) advance it: `beginGesture(ctx)` captures the
generation before any await and exposes `current()`, and
`applySurface(ctx, surface, gesture?)` applies a synchronous gesture or
adopts a captured one only while it is still current; selection actions
accept the same gesture. An older asynchronous gesture (Mission actor or
schedule navigation, task links, new session creation, history-search
navigation, explicit Git actions, browser opens) therefore never
overrides a newer one, while keeping its durable effects (a created
session stays created; a rejected preview releases its own token).
Passive updates (Git status, Mission inventory and passage
reconciliation, restoration and hydration, session/worktree changes,
preview location/history notifications) apply immediately against
current state, never advance the generation and never cancel a pending
gesture. They keep their own guards at the moment they run: a file
reader closes only when a loaded status inventory shows the file is
gone, a Mission reader is cleared only when its Mission is actually
missing, and session deletion reconciles the selected parent and child
at once.

I extract the shared active-session signal into
`stores/session-selection-state.ts`, keeping session actions, catalog
refresh and restoration authority in `stores/session-state.ts` and their
UI-effect dependencies outside the view-state import graph. I add
`tab-view-state.test.ts`, `session-surface.test.ts` and
`app-session-workspace-hydration.test.ts` to the explicit CI unit-test
lists.

I bound late-response fences to in-flight operations and open
occurrences, so repeated open/close cycles retain nothing (1,000-cycle
regression test), while a late success or error for a closed tab still
publishes nothing.

## Validation

I rebased the seven commits onto `dev` at `2541a9195`; `git range-diff`
shows six identical patches and one integrated with the Missions changes
from #1507 (shared passage acknowledgements kept, presentation scoped
per tab).

- At `a4575ac26` I replaced the deferred reconciliation queue from
earlier review rounds with the gesture generation above, reducing the
production diff from +1,446/-703 to +1,227/-690 lines under
`packages/ui/src`. `npm run typecheck` passes; the exact CI step passes
**742/742**; UI source units **2,221** passed (known
`help-url.test.ts:63`) and contract units **190/190**.
- At `22efe6da4` I migrate Mission views saved by earlier builds
(`mission-project-*` selection, reader and expanded rows;
`mission-disclosures-*` choices) into the first tab occurrence without
overriding an existing envelope; the exact CI step passes **745/745**,
and I updated three Mission browser checks to expect immediate passive
reconciliation followed by the pending browser opening.
- At `4c62b5eb1`: `npm run typecheck` passes; the exact CI step passes
**728/728**; UI source units **2,222/2,223** (known
`help-url.test.ts:63`) and contract units **190/190**; the affected
browser files pass **175/175** on Windows and on Linux; the full
Linux/Xvfb browser suite runs **2,133 passed, 2 skipped, 2 failed** (the
two known prompt-menu checks). The Mission recovery and parent/child
deletion regressions fail on `97b23df23` and pass now; a reverse-order
file lifetime/status regression covers stable reader lifetime identity
across passive retargets.
- At `97b23df23`: `npm run typecheck` passes; the exact CI step "Test
changed runnable UI behavior" passes **724/724** on Windows and Linux;
full UI source and contract units pass **2,402/2,403** on both, the
remaining one being the known `help-url.test.ts:63`; the affected
browser files pass **249/249** on Windows; the full Linux/Xvfb browser
suite runs **2,132 passed, 2 skipped, 2 failed** (the two known
prompt-menu checks). The new gesture/passive regressions (12 unit, 3
real-shell) fail on `b099e5a6f` and pass now.
- At `b099e5a6f`: the exact CI step passed **719/719**; full UI units
**2,389 passed** with the known `help-url.test.ts:63` failure; the
affected browser files (current-passage, Mission control,
cross-navigation, task reader, browser history, header windows, current
pages) pass **167/167** on Windows and on Linux; the agent-selector
shortcut check, which I changed to await the visible search popup with
focus and Escape dismissal instead of a transient focus precursor,
passes **10/10** on each platform; the full Linux/Xvfb browser suite
runs **2,129 passed, 2 skipped, 2 failed** (the two known prompt-menu
checks). Each Missions race regression fails on the commit before its
fix and passes now.
- The earlier results below are from previous review rounds, kept as
historical evidence.
- After the second review round, the full Linux/Xvfb browser suite ran
**2,118 passed, 2 skipped, 2 failed** (the two known prompt-menu
checks); the affected browser files ran **116/116** on Windows and on
Linux; full UI unit tests pass on both except the known
`help-url.test.ts:63`.
- After the first review round, the full Linux/Xvfb browser suite ran
**2,106 passed, 2 skipped, 3 failed**: the two known prompt-menu checks,
and a standalone Missions fixture missing tab ownership registration,
which I corrected without changing its assertions (**33/33** on Windows
and Linux afterwards). Electron controller/lifetime/security **29/29**;
help **31/31**.
- New duplicate-tab coverage: a surface transition matrix with
delayed-open races (`session-surface.test.ts`), real localhost clicks,
diff readers and session/Info/history isolation in two tabs, and a
production capture/restore scenario with reorder, duplicate close,
recreated instance ids and no automatic page load
(`tab-view-persistence.test.ts`).
- After the rebase: Missions UI units **347 passed**, tab-state/preview
units **17 passed**, server recurrence **115 passed**, and **373 browser
tests across 20 files** passed, including the duplicate-tab suites
`session-browser-history.test.ts` and `history-navigation.test.ts`.
- Before the rebase, independent QA ran the full UI browser suite on
Windows and on Linux/Xvfb, the full UI unit tests on both, Electron
controller tests and Tauri Rust controller regressions. Duplicate-tab
scenarios run on real Solid components, real Electron guests, the Tauri
IPC boundary and the iframe fallback.
- I tested Tauri at its IPC boundary and through its Rust controller
tests only; I did not run a native WebKit/WebView2 Tauri app.
- Failures I reproduced identically on the previous `dev` base and left
unchanged: `help-url.test.ts:63`, Windows
`workspace-file-refresh.test.ts:67`, and the two Linux `the prompt menu
jumps` variants. Two intermittent Missions browser cases (Linux source
ABA read count, Windows source-picker keyboard scroll) did not reproduce
in 160 comparative runs on either side.

## Documentation

I update `docs/help/<locale>/` `browser-sidecars.md`, `projects.md`,
`files.md`, `missions.md`, `desktop.md` and `settings.md` in all ten
languages, along with `BROWSER_AUTOMATION.md`,
`BROWSER_DEVICE_EMULATION.md`, `UI_SHELL.md`, `PROJECTS.md`,
`FILES_PANEL.md`, `MISSIONS.md` and `architecture.md` under `dev-docs/`.
No new UI strings.

## Long files touched

Under `packages/ui/src/`: `App.tsx` (~910 lines),
`components/browser-frame.tsx` (~1,052),
`components/instance/instance-shell2.tsx` (~1,344),
`components/instance/shell/right-panel/tabs/MissionControl.tsx` (~891),
`components/instance/shell/right-panel/useGitChanges.ts` (~586),
`components/message-block.tsx` (~1,775), `components/mission-reader.tsx`
(~556), `components/session-list.tsx` (~1,165),
`components/session/session-view.tsx` (~731),
`lib/hooks/use-commands.ts` (~683), `stores/session-actions.ts` (~744),
`stores/session-state.ts` (~1,375), `stores/session-api.ts` (~2,074),
`stores/client-state-partitions.ts` (~557). Tests:
`history-navigation.test.ts` (~1,812), `session-browser-history.test.ts`
(~1,029), `header-windows.test.ts` (~1,090),
`mission-current-passage.test.ts` (~807),
`browser_controller_regressions.rs` (~962). Workflow:
`.github/workflows/pr-build.yml` (~1,074).
Trim repetition and historical commentary in developer documentation and
agent guidance without changing contracts or behavior. Areas: runtime
and ownership, Missions, memory, browser and transcript workflows,
settings, repository skills and bundled Design recipes.

Preserve current dev contracts, safety boundaries and agent rules.
Proposed design changes remain outside this editorial pass.

Validated with Design unit and bundle tests, isolated native skill
discovery and loading, palette quality, UI typecheck and help tests.
Align three developer documents with current code: cumulative Changes
reveals and sticky pins, exact palette contrast exceptions, and
persisted mission taskMode requirements. Documentation only. Verified
with 40 palette-quality tests, 10 Design skill-content tests and git
diff --check; the Missions skill has no dedicated content test.
)

Reconcile failed preference writes against the stored choices and offer
read-only catalog Reload using the existing ten-language label. Verified
with failing-first browser regressions, UI typecheck, 699 targeted
unit/i18n/help-contract checks, 42 browser checks and 31 help checks.
Prevent preview capabilities and SideCar proxies from turning outbound connections into privileged local requests through DNS aliases. Refuse backend listener destinations before TCP contact, including sibling HTTP/HTTPS listeners, wildcard-bound interfaces and Remote Control ingress.

Resolve once per connection and supply only checked answers to native HTTP/TLS connectors, preserving Host, TLS identity and IPv4/IPv6 fallback. Cover HTTP previews, capability hosts, Referer assets and both WebSocket proxies through the shared destination boundary; document the restriction in developer docs and all ten help locales.

Validation on Node 24.22.3: red-first regression reproduced 200 with skipped auth and 401 with auth enabled. Server typecheck passes; focused proxy/preview/SideCar/Remote Control/network-policy suites pass 208 tests; CI-exact server glob at concurrency 4 passes 4444 with 25 skipped and no failures; help validation passes 31 tests across 28 pages and ten locales.
Unavailable Missions now opens OpenCode settings for plugin failures or
the existing server diagnostic report for local bridge failures, with
shorter messages. Configuration reload asks for confirmation in the
shared setup panel and names interruption across every connected client;
actions and help cover all ten languages. Verified: UI typecheck, 89
unit checks, 83 browser checks (MissionControl, setup, Info), i18n
parity, multilingual help (28 pages × 10 locales), and diff whitespace.
Distinguish stale revisions, differing deletion requests and pending
controls (including native HTTP 503) in all ten locales and help.
Reconcile the original intent after refusal; keep its request, revision
and cleanup option for explicit retries, requiring a new selection for
changed Missions. Verified failing-first regressions, UI/server
typechecks, 73 server tests, 76 UI/contract/locale/help checks, 117
browser cleanup/editor regressions and test:help. Review found and fixed
the native pending-control status mismatch; its browser regression now
uses the real controller and HTTP route.
Treat unspecified IPv4, IPv6 and mapped destinations as local scope on every backend listening port, even when the listener binds one specific address. Register remote proxy session HTTPS servers with the shared destination guard so previews cannot connect to their bootstrap surface.

Add red-first HTTP/WebSocket regressions for literal and DNS-resolved unspecified destinations, port-isolation checks and a remote HTTPS listener registration regression. Audit all network listen calls: production listeners are the main HTTP/HTTPS application, Remote Control ingress and remote proxy sessions; the remaining calls are test fixtures. Update the adjacent Remote Control boundary documentation.

Validation with Node 24.22.3: server typecheck succeeds; 214 focused tests pass; the reviewer hostile fixture retargeted to this checkout with two refusal assertions corrected passes 11/11. The CI-exact server glob at concurrency 4 passes 4450 tests with 25 skipped, zero failed and zero cancelled (4475 total).
Name the action or reading target, distinguish saturated capacity from
unconfirmed sends, give failed sections their own retry, and separate
validation from authentication in ten locales without changing request
retention or global auth recovery. Verified UI typecheck, 453
unit/locale/help-dependent checks, 31 help checks and targeted
real-browser Missions journeys; correct map labels remain unchanged and
the editor diff is one label line. No help expansion: these transient
states introduce no user workflow or internal-mechanism explanation.
Pin resolved proxy destinations and refuse connections to backend
listeners, including DNS aliases.
Apply the boundary to preview HTTP/WebSockets, asset fallback and
SideCars; retain ordinary dev-server connections.
Verified: server typecheck; 208 focused tests; CI server glob
concurrency 4: 4444 passed, 25 skipped, 0 failed; 31 help tests.
…1545)

Rend les conversations suivies modifiables et disponibles dès la
création immédiate, sans conversion. Conserve les sources et réutilise
la lecture bornée au premier Play et ses identités de reprise. Vérifié :
types UI/serveur, 1 478 tests Missions, unités UI, contrats natifs,
parcours navigateur, parité des dix langues et aide.
Resolve pruning busy-state, briefing-admission and interruption-recovery
assertions through their English translation keys instead of stale prose
from before #1536. Audited the 345 changed or removed English keys
across scripts, tests, desktop hosts and workflows with read-only
review. Verified all 21 affected browser tests, script syntax,
whitespace, the pruning bundle build and the complete isolated native
pruning UI scenario on OpenCode 2.3.0.
Une creation incertaine retrouve sa Mission ou rend le formulaire
uniquement apres refus terminal prouve, sans alerte. Les recus restent
lies a la requete et a sa connexion originale jusqu'a leur consommation,
avec des lectures bornees et sans rejouer la creation. Verification:
typechecks UI et serveur, tests Missions serveur et
UI/contrats/navigateur, parite i18n et test:help.
pascalandr and others added 2 commits November 16, 2026 13:49
Every browser tab of the web client was titled "CodeNomad" (fixed in
`index.html`, never updated), so operators with several servers open in
several tabs could not tell them apart without reading the URL.

## Behavior

- The web client titles its tab `<host> — CodeNomad` from the page's
`location.host`. The port appears only when it is not the scheme's
default, so two servers on one machine also differ.
- On loopback hosts (`localhost`, `*.localhost`, 127/8, `[::1]` and
IPv4-mapped loopback in both notations) the title stays `CodeNomad`, so
single-server use is unchanged.
- Electron and Tauri keep their own titles, since one desktop window can
hold projects from several servers.

## Implementation

The pure helper is `packages/ui/src/lib/web-tab-title.ts`, applied once
at startup in `main.tsx` for the web host. `dev-docs/UI_SHELL.md`
records the rule.

## Validation

`web-tab-title` unit tests (host names, non-default ports, bracketed
IPv6, loopback forms) and the UI typecheck pass.

Closes #1548
…s older (#1538)

Sending a prompt while a reply streams triggers a delivery repair read. The
fetched page replaced unchanged rows wholesale, and OpenCode persists an open
text fragment empty until its ending commits, so the previous reply's tail
disappeared until that ending reached the reducer.

Keep the streamed text or reasoning only for a fragment the page cannot have
committed: the row's last fetched part, empty, on a step open on both sides,
with no ending marker (native metadata for text, time.completed for
reasoning). Committed finals, including shorter or empty replacements, keep
full page authority.

Delivery-repair and rotation-replay regressions run in the CI restore
integration list; the guard is qualified against isolated OpenCode 2.3.0
captures for open and ended text and reasoning fragments.

This branch was successfully deployed

1 active (outdated) deployment
github-pages — 8fe73b75 Deployed Nov 16, 2026 by pascalandr via deploy #1086
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants