diff --git a/.aegis-baseline.tsv b/.aegis-baseline.tsv index a7657db..746be58 100644 --- a/.aegis-baseline.tsv +++ b/.aegis-baseline.tsv @@ -243,11 +243,11 @@ ssrf-localhost ./docs/SESSIONS.md 49960de5880e8c687434170f6476605b8fe4aeb9a28632 ssrf-localhost ./docs/SITE-DEPLOYMENT.md 49960de5880e8c687434170f6476605b8fe4aeb9a28632c7995cf3ba831d9763 1 ssrf-localhost ./docs/decisions/0074-aegis-content-keyed-baseline.md 49960de5880e8c687434170f6476605b8fe4aeb9a28632c7995cf3ba831d9763 1 ssrf-localhost ./docs/decisions/0083-aegis-rebless-parity-matrix.md 49960de5880e8c687434170f6476605b8fe4aeb9a28632c7995cf3ba831d9763 1 -ssrf-localhost ./docs/release/EXECUTION-BOARD.md 49960de5880e8c687434170f6476605b8fe4aeb9a28632c7995cf3ba831d9763 3 +ssrf-localhost ./docs/release/EXECUTION-BOARD.md 49960de5880e8c687434170f6476605b8fe4aeb9a28632c7995cf3ba831d9763 4 ssrf-localhost ./docs/tasks/T22-screenshot-ui.md 49960de5880e8c687434170f6476605b8fe4aeb9a28632c7995cf3ba831d9763 1 ssrf-localhost ./docs/tasks/T27-12-adjudicate-aegis-findings.md 12ca17b49af2289436f303e0166030a21e525d266e209267433801a8fd4071a0 1 ssrf-localhost ./docs/tasks/T27-12-adjudicate-aegis-findings.md 49960de5880e8c687434170f6476605b8fe4aeb9a28632c7995cf3ba831d9763 7 -ssrf-localhost ./docs/tasks/T27-24-pr8-remaining-findings.md 49960de5880e8c687434170f6476605b8fe4aeb9a28632c7995cf3ba831d9763 2 +ssrf-localhost ./docs/tasks/T27-24-pr8-remaining-findings.md 49960de5880e8c687434170f6476605b8fe4aeb9a28632c7995cf3ba831d9763 5 ssrf-localhost ./packaging/deb/control 49960de5880e8c687434170f6476605b8fe4aeb9a28632c7995cf3ba831d9763 1 ssrf-localhost ./scripts/install-aegis-selftest.sh 12ca17b49af2289436f303e0166030a21e525d266e209267433801a8fd4071a0 2 ssrf-localhost ./scripts/reconcile_deployment-selftest.py 12ca17b49af2289436f303e0166030a21e525d266e209267433801a8fd4071a0 5 @@ -744,7 +744,7 @@ zip-code ./docs/tasks/T27-24-pr8-remaining-findings.md 36b212ef149cc79e63eb10917 zip-code ./docs/tasks/T27-24-pr8-remaining-findings.md 6a0d62bf9572742690225db40eb2eb3b89fd0574bcadf54e48f4f71b1b090977 1 zip-code ./docs/tasks/T27-24-pr8-remaining-findings.md 8f0287fed25004a3261c7953e4c8286dd2f639c21b55323b7c931781c6f6abcb 1 zip-code ./docs/tasks/T27-24-pr8-remaining-findings.md b3cb7d030fce5d1824bae203e63a8455fbf0b817f8b5d9a2174c704179204471 1 -zip-code ./docs/tasks/T27-24-pr8-remaining-findings.md dabf5c8b9082180e7d7ae1ee52a9417fc76be7e3e80aed2a8f3cccf1cd1c162d 3 +zip-code ./docs/tasks/T27-24-pr8-remaining-findings.md dabf5c8b9082180e7d7ae1ee52a9417fc76be7e3e80aed2a8f3cccf1cd1c162d 4 zip-code ./docs/tasks/T27-41-windows-arm64-assessment.md 36b212ef149cc79e63eb109177f831812969fa8e86f9c09b405448276b0ec76f 1 zip-code ./docs/tasks/T27-41-windows-arm64-assessment.md 6a0d62bf9572742690225db40eb2eb3b89fd0574bcadf54e48f4f71b1b090977 2 zip-code ./docs/tasks/T27-41-windows-arm64-assessment.md 8f0287fed25004a3261c7953e4c8286dd2f639c21b55323b7c931781c6f6abcb 1 diff --git a/.aegisignore b/.aegisignore index b7c5b0b..c366634 100644 --- a/.aegisignore +++ b/.aegisignore @@ -53,6 +53,11 @@ temp/ # them, so their findings can never stick to the baseline. .gitignore-scratch/ +# Session scratch: helper scripts, renumbered document copies and probe +# artifacts. Git-ignored working files, not source; their content churns +# with every session, so their findings can never stick to the baseline. +.scratch/ + # Proprietary Adobe reference install (git-ignored working material, # see CLAUDE.md source/ rules). Third-party binaries and vendor JS — # every finding there is noise by definition, and the files churn. diff --git a/.gitignore b/.gitignore index 0408dc7..c16d854 100644 --- a/.gitignore +++ b/.gitignore @@ -36,6 +36,11 @@ lcov.info # Session scratch: throwaway helper scripts and probe artifacts /.claude/scratch/ +/.scratch/ + +# Worktree-local build scratch (a private cargo target dir; the shared +# /cargo-target tree belongs to another session) +/.cargo/ # Python bytecode from the offline pipeline validator __pycache__/ diff --git a/docs/TASKS.md b/docs/TASKS.md index a2ff531..4fae8d9 100644 --- a/docs/TASKS.md +++ b/docs/TASKS.md @@ -79,7 +79,7 @@ Review follow-up: [filter controls and master transform](tasks/filter-controls-r | — | CI false red: a job compiles against an `app` artifact built from another tree (`E0425: cannot find function 'menu_census' in module 'app::menus'`) | [HANDOFF §6.1](HANDOFF.md#61-pipeline) | `OPEN — DEFECT, NOT A SOURCE FAULT` — measured 2026-11-16 on six branch runs (`a329bda0`, `e2e85b9b`, `57660fb8`, `5e48f383`, `b28fd3f5`, `0757082e`, `a6b23df6`); each run's own commit carries both the definition and the call, so the tree was consistent and the artifact was not. The mechanism is not yet identified: this pipeline declares no `cache:` mount, the CI image bakes no target directory, and the runner creates a fresh container per workspace job | **GitForge CI** (not this repository) | identify where the stale `app` rlib enters the job — the runner's artifact handling is the lead — and fix it there; until then, a run that fails with this signature is re-triggered (an empty commit) rather than read as a source defect | | — | Release workflow hardening: every action pinned to a commit SHA, a fail-closed pin check in CI (decision 0099), a classified release mode with a fail-closed aggregate gate (decision 0100), a pinned measuring toolchain (decision 0102) | [decisions/0099-workflow-actions-are-pinned.md](decisions/0099-workflow-actions-are-pinned.md) · [decisions/0100-release-mode-classification-and-a-tested-release-gate.md](decisions/0100-release-mode-classification-and-a-tested-release-gate.md) · [decisions/0102-coverage-floor-is-measured-with-a-pinned-toolchain.md](decisions/0102-coverage-floor-is-measured-with-a-pinned-toolchain.md) | `IMPLEMENTED` — the 19 action references are pinned; the aggregate gate is a tested script that refuses a missing, empty, extra, mismatched or platform-less artifact set; a rehearsal is a `TEST`-titled prerelease on a disposable tag; the validator and the three release self-tests pass locally; the CI receipt is pending | CI/tooling lane | land the branch on both receipts; the checks run in the backup CI's `changes` job, and mirroring them into the GitForge pipeline waits for the `:2` pipeline shape (PR #123, PR #129); then rehearse the release path end to end through `workflow_dispatch` | | — | The coverage gate explains itself: the per-file breakdown, each file's `DA` record count beside its `LF`, and the profile inventory (decision 0103) | [decisions/0103-coverage-export-counts-a-file-once-per-build.md](decisions/0103-coverage-export-counts-a-file-once-per-build.md) | `IMPLEMENTED` on the branch; the CI receipt is pending | CI/tooling lane | land it on both receipts; the check that closes the decision is the next pinned run, where `LF` must equal the `DA` count again for `crates/app/src/shell/mod.rs` — under rustc 1.100.0 the export reported 8138 lines for a file that has 6309 | -| 27 | Cross-platform release engineering: Windows, macOS, Linux package matrix, Android feasibility, source, SBOM/provenance, signing, CI aggregation and certification | [tasks/T27-cross-platform-release.md](tasks/T27-cross-platform-release.md) · the atomic ledger is [release/EXECUTION-BOARD.md](release/EXECUTION-BOARD.md) (62 rows, one per `docs/tasks/T27-*.md`) | `IN PROGRESS` — implementation is ahead of verification; **no target is `CERTIFIED`** and nothing is signed; the SBOM half is build-verified against the vendored CycloneDX 1.5 schema and compared against the lockfile's own inventory (decision [0119](decisions/0119-the-sbom-is-validated-against-the-schema-it-claims.md)) | release-workflow lane | land an end-to-end tag run of the repaired workflow and attach the run id as the RP-07 receipt; consolidate the duplicated `dist --source`/`--sbom` verbs (B-3); the board ledger was reconciled on 2026-11-16 (T27-59) — 62 rows against 62 tickets, and one live application defect surfaced from PR #8's undispositioned comments is now carried by T27-24 | +| 27 | Cross-platform release engineering: Windows, macOS, Linux package matrix, Android feasibility, source, SBOM/provenance, signing, CI aggregation and certification | [tasks/T27-cross-platform-release.md](tasks/T27-cross-platform-release.md) · the atomic ledger is [release/EXECUTION-BOARD.md](release/EXECUTION-BOARD.md) (62 rows, one per `docs/tasks/T27-*.md`) | `IN PROGRESS` — implementation is ahead of verification; **no target is `CERTIFIED`** and nothing is signed; the SBOM half is build-verified against the vendored CycloneDX 1.5 schema and compared against the lockfile's own inventory (decision [0119](decisions/0119-the-sbom-is-validated-against-the-schema-it-claims.md)) | release-workflow lane | land an end-to-end tag run of the repaired workflow and attach the run id as the RP-07 receipt; consolidate the duplicated `dist --source`/`--sbom` verbs (B-3); the board ledger was reconciled on 2026-11-16 (T27-59) and again on 2026-11-17 — 62 rows against 62 tickets, no problems; PR #8's undispositioned comments are adjudicated (T27-24) and the one valid application defect among them — a plotted draft carrying no owning tool — is fixed on `main` by `d3e45cf9` (T27-23, PR #171), and the Aegis scan now runs on documentation-only changes in both CI systems (T27-62, PR #172 → `7bcf28fd`) | | 26 | Refresh user-visible CS6 parity matrix | [tasks/T26-parity-matrix.md](tasks/T26-parity-matrix.md) | `IMPLEMENTED` — matrix published | parity lane | every CS6 menu, tool, panel, inspector field set, workflow, import/export surface, canvas interaction, keymap area, and selection state is classified with an evidence citation and an owning ticket in [`CS6-PARITY.md`](CS6-PARITY.md); the browser-sampling check the ticket named was performed on 2026-11-04 against a fresh optimized bundle and is recorded in T22 | | 22 | Phase 22: CS6 shell fidelity and screenshot audit | [tasks/T22-screenshot-ui.md](tasks/T22-screenshot-ui.md) | `IN PROGRESS` | UI lane | screenshot acceptance is complete; 22.1 panel inventory and 22.2 tool strip landed on `main` with the shell-lineage train (#107); 22.5 chrome remains partial, and 22.4 still has evidence-gated bitmap transform/fill-stroke work. [Phase 22](NEXT.md#phase-22--cs6-shell-fidelity-p0-the-looks-and-acts-like-fireworks-phase) | | 20 | 20.1 In-canvas text editing (egui TextEdit overlay; kittest focus tests) | [tasks/T20-text-editing.md](tasks/T20-text-editing.md) | `IMPLEMENTED` — landed 2026-11-06 (`shell::begin_text_edit` / `text_edit_overlay`) | editing lane | Phase 20's other items are tracked in the rows above and in the Phase 20 board below | diff --git a/docs/release/EXECUTION-BOARD.md b/docs/release/EXECUTION-BOARD.md index 967123e..8a57305 100644 --- a/docs/release/EXECUTION-BOARD.md +++ b/docs/release/EXECUTION-BOARD.md @@ -170,7 +170,7 @@ condition/verification. **RP** is the release-plan work package the row belongs | T27-15 | Get the authoritative GitForge pipeline green on the same SHA | P0 | IN PROGRESS | T27-13 | The `firecraft-ci` pipeline concludes success on the same commit as T27-14 | GitForge API status call for the run id of the pushed SHA | GitForge runs exist for all six candidate heads, but the fleet has one online runner (`swarmone-docker`, capacity 4) and the candidate queue exceeds it, so the runs are still in flight. The fleet measurement and the two GitForge-side defects — a superseded run's running job keeps its runner slot, and the CLI has no cancel verb — are recorded in issue #126. No green GitForge run is claimed until the authoritative store reports `succeeded`. | the fleet's single online runner; T27-13 is closed | — | | T27-16 | Make the release path run the web bridge smoke with no skip | P0 | BUILD VERIFIED | T27-05 | The release path executes the bridge smoke; a skip exists only for non-release CI and is documented as such | `grep -n FIRECRAFT_DIST_SKIP_SMOKE .gitforge.yml .github/workflows/*.yml` | The row's premise was wrong and is corrected here. `FIRECRAFT_DIST_SKIP_SMOKE` appears nowhere in `.gitforge.yml` or the GitHub workflows on `main` (it survives only in [`xtask/src/dist.rs`](https://github.com/aliasfoxkde/FireCraft/blob/main/xtask/src/dist.rs) (line 52) as a comment, in the docs, and in issue #110), and `infrastructure/docker/ci-rust.Dockerfile:127` installs bookworm `chromium` for exactly this smoke. The smoke runs in CI: the GitForge `release` job logs `[xtask] dist web: 5 file(s), wasm 15573974 B (warning 25000000, hard ceiling 50000000), manifest MANIFEST.sha256, smoke green` and `[xtask] dist --web ok` — most recently job `ecb488d1-8965-443f-8309-71405883d754` for commit `fcfe6db1`, finished 2026-11-12T06:18:54Z, and 12 release jobs carry that line. | — | 2026-11-12: smoke green in the GitForge release job | | T27-59 | Keep the execution board and the ticket ledger truthful as rows close | P1 | IMPLEMENTED | T27-07 | Every closed row carries a completion date and a cited evidence link; no row claims a status stronger than its evidence; the board's task-ID set and `docs/tasks/T27-*.md` stay in one-to-one correspondence | `bash scripts/verify_release_ledger.sh .` | the board's §6 table and the 62 ticket files now state the same status, evidence, blockers and completion date for every row; [`scripts/verify_release_ledger.sh`](https://github.com/aliasfoxkde/FireCraft/blob/main/scripts/verify_release_ledger.sh) reports `rows=62 tickets=62 problems=0`. The reconciliation is a tracked tool, not a one-off script: `scripts/reconcile_release_ledger.py` writes the board's cells into their tickets (the board is the only source of truth), and the verifier is the gate that keeps them in step. The verifier also refuses a closed row that cites no evidence link (NO LINK) — the board's acceptance criterion is a completion date *and* a cited evidence link, so every advanced row now carries one and the gate keeps it that way. The four rows whose verification cell carries an escaped pipe inside a code span — T27-08, T27-22, T27-41 and this one — keep the escape, so each row is ten cells and not eleven. | — | 2026-11-15: the board and its 61 tickets reconciled and the verifier landed with a CI step | -| T27-62 | Run the Aegis secret scan on documentation-only changes, in both CI systems | P1 | IN PROGRESS | T27-24 | A documentation-only change is scanned by the Aegis leg in both CI systems; the scanner is installed from one script with one pinned checksum; no CI system skips the scan for a documentation-only diff | `scripts/validate_gitforge_pipeline.py .gitforge.yml && python3 scripts/validate_workflow_pins.py .github/workflows` | The defect and its fix are in [decision 0132](../decisions/0132-the-secret-scan-runs-on-documentation-changes.md). The scan covers `docs/`, but both CI systems ran it only for code changes: the GitHub `gate` job is gated on `code_changed` and its `docs-only-check` job validated whitespace only, and the GitForge pipeline had no Aegis job at all while every Cargo job carries a `skip_if` on `CI_CHANGE_INCLUDE_CODE`. PR #170's documentation-only merge therefore entered `main` with four unblessed Windows build literals, and the next code change's gate failed for content it never touched: on `main` `c26adffc` the `gate` job's Aegis leg reported 5 findings the baseline does not cover. The same defect is visible from the other CI system: PR #154 (`8067b53b`, the icon-resource survey) touches no file under `docs/tasks/`, and its GitForge run `fe161636` **succeeded** because the pipeline has no Aegis job to answer, while the GitHub `gate` job on the same SHA **failed** with the same 5 findings, every one of them in `docs/tasks/T27-41-windows-arm64-assessment.md`. The fix has three parts. `scripts/install-aegis.sh` is the single installer — one release URL, one SHA-256, one identity check (`aegis --help` must list the `scan` subcommand, because an unrelated tool also ships under the bare name `aegis`) — and it brings its own fetch, because the authoritative CI image ships no `curl` and no `wget`; the GitHub `gate` job's inline installer now calls it instead of carrying its own copy of the pin. The GitHub `docs-only-check` job installs Rust and the scanner and runs `cargo run --package xtask -- aegis`. The GitForge pipeline gains an `aegis` job with **no `skip_if`**, chained between `release-ledger` and `fmt-layers`, with its own cache directory. The pipeline definition validates (`scripts/validate_gitforge_pipeline.py`: 12 jobs, one root, no join) and the workflow's action pins validate. | — | — | +| T27-62 | Run the Aegis secret scan on documentation-only changes, in both CI systems | P1 | IMPLEMENTED | T27-24 | A documentation-only change is scanned by the Aegis leg in both CI systems; the scanner is installed from one script with one pinned checksum; no CI system skips the scan for a documentation-only diff | `scripts/validate_gitforge_pipeline.py .gitforge.yml && python3 scripts/validate_workflow_pins.py .github/workflows` | The defect and its fix are in [decision 0132](../decisions/0132-the-secret-scan-runs-on-documentation-changes.md). The scan covers `docs/`, but both CI systems ran it only for code changes: the GitHub `gate` job is gated on `code_changed` and its `docs-only-check` job validated whitespace only, and the GitForge pipeline had no Aegis job at all while every Cargo job carries a `skip_if` on `CI_CHANGE_INCLUDE_CODE`. PR #170's documentation-only merge therefore entered `main` with four unblessed Windows build literals, and the next code change's gate failed for content it never touched: on `main` `c26adffc` the `gate` job's Aegis leg reported 5 findings the baseline does not cover. The same defect is visible from the other CI system: PR #154 (`8067b53b`, the icon-resource survey) touches no file under `docs/tasks/`, and its GitForge run `fe161636` **succeeded** because the pipeline has no Aegis job to answer, while the GitHub `gate` job on the same SHA **failed** with the same 5 findings, every one of them in `docs/tasks/T27-41-windows-arm64-assessment.md`. The fix has three parts. `scripts/install-aegis.sh` is the single installer — one release URL, one SHA-256, one identity check (`aegis --help` must list the `scan` subcommand, because an unrelated tool also ships under the bare name `aegis`) — and it brings its own fetch, because the authoritative CI image ships no `curl` and no `wget`; the GitHub `gate` job's inline installer now calls it instead of carrying its own copy of the pin. The GitHub `docs-only-check` job installs Rust and the scanner and runs `cargo run --package xtask -- aegis`. The GitForge pipeline gains an `aegis` job with **no `skip_if`**, chained between `release-ledger` and `fmt-layers`, with its own cache directory. The pipeline definition validates (`scripts/validate_gitforge_pipeline.py`: 12 jobs, one root, no join) and the workflow's action pins validate. The fix is merged as PR #172 → `7bcf28fd`, and the scan has since run green on a documentation-only change in **both** systems: the GitForge `aegis` job has no `skip_if` and the GitHub `docs-only-check` job installs the scanner and runs it. | — | 2026-11-17 | ### RP-01 — Linux package completion @@ -257,9 +257,9 @@ condition/verification. **RP** is the release-plan work package the row belongs | ID | Title | Pri | Status | Depends on | Acceptance criteria | Verification | Evidence | Blockers | Done | |---|---|---|---|---|---|---|---|---|---| -| T27-24 | Verify the remaining PR #8 findings, fix the valid ones and document the rejected ones | P1 | IN PROGRESS | T27-03 | Every remaining Major/Critical finding is independently verified against current code; valid ones are fixed; rejected ones carry a reason | `cargo test -p app -p engine` | All 9 code findings and 25 documentation findings in PR #8's inline comments were adjudicated against `main` `f1a45bec`. Every code finding is already correct on `main`, so each is rejected with the code that is right: `editor.rs:619` `drag_quadrangle` moves the grabbed corner and the next adjacent corner (`let adjacent = (grabbed + 1) % 4;`), which preserves `TL + BR = TR + BL` and matches its doc comment; `editor.rs:2376` Scale, Skew and Distort read `self.live_selection()` (`editor.rs:2494`, `:2525`, `:2563`, `:2587`), not `document().object_selection`; `editor.rs:2595` Redraw Path sends the sampled pointer trace (`self.redraw_selected_path(&drag.trace)` at `editor.rs:2614`; the signature takes `trace: &[geom::Point]` at `:2733`), guarded by `redraw_path_uses_the_sampled_pointer_trace`; `panels_web.rs:595` the `symbol.createInstance` payload carries only `symbol`, `x`, `y`, `page`, `layer` and `cell`, so `CreateInstancePayload`'s `deny_unknown_fields` is satisfied; `panels_web.rs:574` the placement is `symbol_insert_translation` (`panels_web.rs:275`), which centres the symbol's `definition_content_bounds` on the canvas, guarded by the test at `panels_web.rs:1522`; `retouch.rs:656` the four retouch commands call `brush_diameter` (`retouch.rs:118`), which `checked_mul(2)` and returns `EngineError::BadPayload`; `retouch.rs:1149` `retouch_commands_reject_overflowing_brush_diameters` asserts `Err(EngineError::BadPayload { .. })` for all four; `xtask/src/main.rs:222` `main` carries no `install_pattern_state`, no `temp_dir()` and no `firecraft-aegis-home` — the Aegis leg runs the scanner over the repo root against the content-keyed `.aegis-baseline.tsv` (decision 0074); and the Vector Path node drag that a pure click never ends is fixed on `main` by `79fe2759` (#168), which opens the drag in the press routing (`canvas_ui.rs:1060` and `:1111` → `begin_node_reshape`) and closes it on `drag_stopped` (`canvas_ui.rs:1193` → `:1205`), so no transaction leaks. The one valid code finding, `editor.rs:3271-3276` draft tool ownership, is fixed on the T27-23 branch with three regression tests. Of the 25 documentation findings, 23 were already resolved on `main`: the trademark notice at `README.md:22`, `-p app-cli` at `README.md:102` and `:112`, the live-ledger pointer at `README.md:55`, the historical-gaps label at `README.md:156`, the six-cell `1280×800` row and the reconciled menu-resource evidence in `docs/CS6-PARITY.md`, the panel-status corrections in `docs/NEXT.md`, both optimized builds in `docs/PROGRESS.md`, the `ToolPalette`/`Slices` and `57651` evidence notes in `docs/research/fw-ui-evidence.md`, the remaining-work list in `docs/HANDOFF.md`, the L4 crate row and the right-side-dock description in `docs/wiki/`, the closed menu burn-down in `docs/TASKS.md`, the Aegis baseline re-blessed for decision 0083, and the conflict-free `docs/HANDOFF.md:3` orientation. Two remained valid and are fixed here: `docs/wiki/Parity-Roadmap.md` now states that the 2-inert-of-298 census is not a coverage number and names the 35 `ShellAction::ReportGap` rows, and `README.md:60` labels `docs/GAPS.md` the historical gaps audit instead of current gaps. The per-finding table is in [`docs/tasks/T27-24-pr8-remaining-findings.md`](https://github.com/aliasfoxkde/FireCraft/blob/main/docs/tasks/T27-24-pr8-remaining-findings.md). The same change repairs the Aegis baseline and closes the gap that let it drift: PR #170 merged `docs/tasks/T27-41-windows-arm64-assessment.md` through a documentation-only path, and both CI systems skip the Aegis leg for documentation-only changes — GitHub's `gate` job is gated on `code_changed` while its `docs-only-check` job validates whitespace only, and the GitForge pipeline has no Aegis job at all — so four Windows build literals entered `main` unblessed, and the next code change's gate failed for content it never touched (`main` `c26adffc`: 5 findings the baseline does not cover). The literals are adjudicated as false positives in the `zip-code` class note (`xtask/src/aegis.rs`), the baseline is re-blessed (18 keys added, each with a reason in the table below), and the scan now runs on documentation-only changes in both CI systems — [decision 0132](../decisions/0132-the-secret-scan-runs-on-documentation-changes.md), row T27-62. | — | — | +| T27-24 | Verify the remaining PR #8 findings, fix the valid ones and document the rejected ones | P1 | IMPLEMENTED | T27-03 | Every remaining Major/Critical finding is independently verified against current code; valid ones are fixed; rejected ones carry a reason | `cargo test -p app -p engine` | All 9 code findings and 25 documentation findings in PR #8's inline comments were adjudicated against `main` `f1a45bec`. Every code finding is already correct on `main`, so each is rejected with the code that is right: `editor.rs:619` `drag_quadrangle` moves the grabbed corner and the next adjacent corner (`let adjacent = (grabbed + 1) % 4;`), which preserves `TL + BR = TR + BL` and matches its doc comment; `editor.rs:2376` Scale, Skew and Distort read `self.live_selection()` (`editor.rs:2494`, `:2525`, `:2563`, `:2587`), not `document().object_selection`; `editor.rs:2595` Redraw Path sends the sampled pointer trace (`self.redraw_selected_path(&drag.trace)` at `editor.rs:2614`; the signature takes `trace: &[geom::Point]` at `:2733`), guarded by `redraw_path_uses_the_sampled_pointer_trace`; `panels_web.rs:595` the `symbol.createInstance` payload carries only `symbol`, `x`, `y`, `page`, `layer` and `cell`, so `CreateInstancePayload`'s `deny_unknown_fields` is satisfied; `panels_web.rs:574` the placement is `symbol_insert_translation` (`panels_web.rs:275`), which centres the symbol's `definition_content_bounds` on the canvas, guarded by the test at `panels_web.rs:1522`; `retouch.rs:656` the four retouch commands call `brush_diameter` (`retouch.rs:118`), which `checked_mul(2)` and returns `EngineError::BadPayload`; `retouch.rs:1149` `retouch_commands_reject_overflowing_brush_diameters` asserts `Err(EngineError::BadPayload { .. })` for all four; `xtask/src/main.rs:222` `main` carries no `install_pattern_state`, no `temp_dir()` and no `firecraft-aegis-home` — the Aegis leg runs the scanner over the repo root against the content-keyed `.aegis-baseline.tsv` (decision 0074); and the Vector Path node drag that a pure click never ends is fixed on `main` by `79fe2759` (#168), which opens the drag in the press routing (`canvas_ui.rs:1060` and `:1111` → `begin_node_reshape`) and closes it on `drag_stopped` (`canvas_ui.rs:1193` → `:1205`), so no transaction leaks. The one valid code finding, `editor.rs:3271-3276` draft tool ownership, is fixed on `main` by `d3e45cf9` (PR #171) with three regression tests. Of the 25 documentation findings, 23 were already resolved on `main`: the trademark notice at `README.md:22`, `-p app-cli` at `README.md:102` and `:112`, the live-ledger pointer at `README.md:55`, the historical-gaps label at `README.md:156`, the six-cell `1280×800` row and the reconciled menu-resource evidence in `docs/CS6-PARITY.md`, the panel-status corrections in `docs/NEXT.md`, both optimized builds in `docs/PROGRESS.md`, the `ToolPalette`/`Slices` and `57651` evidence notes in `docs/research/fw-ui-evidence.md`, the remaining-work list in `docs/HANDOFF.md`, the L4 crate row and the right-side-dock description in `docs/wiki/`, the closed menu burn-down in `docs/TASKS.md`, the Aegis baseline re-blessed for decision 0083, and the conflict-free `docs/HANDOFF.md:3` orientation. Two remained valid and are fixed here: `docs/wiki/Parity-Roadmap.md` now states that the 2-inert-of-298 census is not a coverage number and names the 35 `ShellAction::ReportGap` rows, and `README.md:60` labels `docs/GAPS.md` the historical gaps audit instead of current gaps. The per-finding table is in [`docs/tasks/T27-24-pr8-remaining-findings.md`](https://github.com/aliasfoxkde/FireCraft/blob/main/docs/tasks/T27-24-pr8-remaining-findings.md). The same change repairs the Aegis baseline and closes the gap that let it drift: PR #170 merged `docs/tasks/T27-41-windows-arm64-assessment.md` through a documentation-only path, and both CI systems skip the Aegis leg for documentation-only changes — GitHub's `gate` job is gated on `code_changed` while its `docs-only-check` job validates whitespace only, and the GitForge pipeline has no Aegis job at all — so four Windows build literals entered `main` unblessed, and the next code change's gate failed for content it never touched (`main` `c26adffc`: 5 findings the baseline does not cover). The literals are adjudicated as false positives in the `zip-code` class note (`xtask/src/aegis.rs`), the baseline is re-blessed (24 keys added — 18 `zip-code`, 4 `ssrf`, 2 `ssrf-localhost` — each with a reason in the table below), and the scan now runs on documentation-only changes in both CI systems — [decision 0132](../decisions/0132-the-secret-scan-runs-on-documentation-changes.md), row T27-62. | — | 2026-11-17 | | T27-25 | Run the full application suite, the CI gate and the web smoke for PR #8 | P0 | NOT STARTED | T27-23, T27-24 | The app test suite, `xtask ci`, the web build and the browser smoke all pass on the PR head | `cargo run -p xtask -- ci && cargo run -p xtask -- dist --web` | none yet — PR #8 was closed unmerged; its review findings are re-hosted in issue #47 and the parity gate in closed issue #86. | T27-23, T27-24 | — | -| T27-23 | Track the originating tool in plotted-draft state, and cover it with regression tests | P0 | IN PROGRESS | T27-03 | Draft state records its owning plotting tool; switching tools resets or explicitly converts the draft; regression tests cover Pen→Hotspot and Hotspot→Pen | `cargo test -p app` | CodeRabbit Major on `crates/app/src/editor.rs:3271-3276` (Pen points folded into a Hotspot commit) — reproduced against the current code and fixed on branch `fix/plotted-draft-keeps-its-tool`: the pending draft now records the tool that plotted it (`PointDraft::tool`), a tool switch converts the pending path explicitly (committed as the tool that plotted it, dropped when it has too few points to become an object), and `commit_draft` dispatches on the draft's own tool rather than the selected one. The regressions are in [`crates/app/tests/plotted_draft.rs`](https://github.com/aliasfoxkde/FireCraft/blob/fix/plotted-draft-keeps-its-tool/crates/app/tests/plotted_draft.rs) — Pen→Hotspot, Hotspot→Pen and the too-few-points drop. | — | — | +| T27-23 | Track the originating tool in plotted-draft state, and cover it with regression tests | P0 | IMPLEMENTED | T27-03 | Draft state records its owning plotting tool; switching tools resets or explicitly converts the draft; regression tests cover Pen→Hotspot and Hotspot→Pen | `cargo test -p app` | CodeRabbit Major on `crates/app/src/editor.rs:3271-3276` (Pen points folded into a Hotspot commit) — reproduced against the current code and fixed: the pending draft now records the tool that plotted it (`PointDraft::tool`), a tool switch converts the pending path explicitly (committed as the tool that plotted it, dropped when it has too few points to become an object), and `commit_draft` dispatches on the draft's own tool rather than the selected one. The regressions are in [`crates/app/tests/plotted_draft.rs`](https://github.com/aliasfoxkde/FireCraft/blob/main/crates/app/tests/plotted_draft.rs) — Pen→Hotspot, Hotspot→Pen and the too-few-points drop. Merged as PR #171 → `d3e45cf9`; the GitHub `gate`, `fmt` and `classify-change` checks are green on that head and the GitForge pipeline for it (`93905d80`, head `7ff1321e`) succeeded. | — | 2026-11-17 | ### PR #8 finding census (all PR #8 inline comments, for T27-23/24/25) diff --git a/docs/tasks/T27-23-draft-tool-ownership.md b/docs/tasks/T27-23-draft-tool-ownership.md index 91dbbbe..def9a18 100644 --- a/docs/tasks/T27-23-draft-tool-ownership.md +++ b/docs/tasks/T27-23-draft-tool-ownership.md @@ -1,6 +1,6 @@ # T27-23 — Track the originating tool in plotted-draft state, and cover it with regression tests -> **RP:** Phase-2 (PR #8 remediation) · **Priority:** P0 · **Status:** IN PROGRESS +> **RP:** Phase-2 (PR #8 remediation) · **Priority:** P0 · **Status:** IMPLEMENTED > **Board row:** [EXECUTION-BOARD.md §6](../release/EXECUTION-BOARD.md) > **Depends on:** T27-03 @@ -19,7 +19,7 @@ Hotspot → Pen, and fail against the pre-fix behaviour. **Verification:** `cargo test -p app` -**Evidence:** CodeRabbit Major on `crates/app/src/editor.rs:3271-3276` (Pen points folded into a Hotspot commit) — reproduced against the current code and fixed on branch `fix/plotted-draft-keeps-its-tool`: the pending draft now records the tool that plotted it (`PointDraft::tool`), a tool switch converts the pending path explicitly (committed as the tool that plotted it, dropped when it has too few points to become an object), and `commit_draft` dispatches on the draft's own tool rather than the selected one. The regressions are in [`crates/app/tests/plotted_draft.rs`](https://github.com/aliasfoxkde/FireCraft/blob/fix/plotted-draft-keeps-its-tool/crates/app/tests/plotted_draft.rs) — Pen→Hotspot, Hotspot→Pen and the too-few-points drop. +**Evidence:** CodeRabbit Major on `crates/app/src/editor.rs:3271-3276` (Pen points folded into a Hotspot commit) — reproduced against the current code and fixed: the pending draft now records the tool that plotted it (`PointDraft::tool`), a tool switch converts the pending path explicitly (committed as the tool that plotted it, dropped when it has too few points to become an object), and `commit_draft` dispatches on the draft's own tool rather than the selected one. The regressions are in [`crates/app/tests/plotted_draft.rs`](https://github.com/aliasfoxkde/FireCraft/blob/main/crates/app/tests/plotted_draft.rs) — Pen→Hotspot, Hotspot→Pen and the too-few-points drop. Merged as PR #171 → `d3e45cf9`; the GitHub `gate`, `fmt` and `classify-change` checks are green on that head and the GitForge pipeline for it (`93905d80`, head `7ff1321e`) succeeded. **Done:** 2026-11-17. **Blockers:** — diff --git a/docs/tasks/T27-24-pr8-remaining-findings.md b/docs/tasks/T27-24-pr8-remaining-findings.md index 2eba275..f59beac 100644 --- a/docs/tasks/T27-24-pr8-remaining-findings.md +++ b/docs/tasks/T27-24-pr8-remaining-findings.md @@ -1,6 +1,6 @@ # T27-24 — Verify the remaining PR #8 findings, fix the valid ones, document the rejected ones -> **RP:** Phase-2 (PR #8 remediation) · **Priority:** P1 · **Status:** IN PROGRESS +> **RP:** Phase-2 (PR #8 remediation) · **Priority:** P1 · **Status:** IMPLEMENTED > **Board row:** [EXECUTION-BOARD.md §6](../release/EXECUTION-BOARD.md) > **Depends on:** T27-03 @@ -22,7 +22,7 @@ ones carry a written reason naming the code that is actually correct. **Verification:** `cargo test -p app -p engine` -**Evidence:** All 9 code findings and 25 documentation findings in PR #8's inline comments were adjudicated against `main` `f1a45bec`. Every code finding is already correct on `main`, so each is rejected with the code that is right: `editor.rs:619` `drag_quadrangle` moves the grabbed corner and the next adjacent corner (`let adjacent = (grabbed + 1) % 4;`), which preserves `TL + BR = TR + BL` and matches its doc comment; `editor.rs:2376` Scale, Skew and Distort read `self.live_selection()` (`editor.rs:2494`, `:2525`, `:2563`, `:2587`), not `document().object_selection`; `editor.rs:2595` Redraw Path sends the sampled pointer trace (`self.redraw_selected_path(&drag.trace)` at `editor.rs:2614`; the signature takes `trace: &[geom::Point]` at `:2733`), guarded by `redraw_path_uses_the_sampled_pointer_trace`; `panels_web.rs:595` the `symbol.createInstance` payload carries only `symbol`, `x`, `y`, `page`, `layer` and `cell`, so `CreateInstancePayload`'s `deny_unknown_fields` is satisfied; `panels_web.rs:574` the placement is `symbol_insert_translation` (`panels_web.rs:275`), which centres the symbol's `definition_content_bounds` on the canvas, guarded by the test at `panels_web.rs:1522`; `retouch.rs:656` the four retouch commands call `brush_diameter` (`retouch.rs:118`), which `checked_mul(2)` and returns `EngineError::BadPayload`; `retouch.rs:1149` `retouch_commands_reject_overflowing_brush_diameters` asserts `Err(EngineError::BadPayload { .. })` for all four; `xtask/src/main.rs:222` `main` carries no `install_pattern_state`, no `temp_dir()` and no `firecraft-aegis-home` — the Aegis leg runs the scanner over the repo root against the content-keyed `.aegis-baseline.tsv` (decision 0074); and the Vector Path node drag that a pure click never ends is fixed on `main` by `79fe2759` (#168), which opens the drag in the press routing (`canvas_ui.rs:1060` and `:1111` → `begin_node_reshape`) and closes it on `drag_stopped` (`canvas_ui.rs:1193` → `:1205`), so no transaction leaks. The one valid code finding, `editor.rs:3271-3276` draft tool ownership, is fixed on the T27-23 branch with three regression tests. Of the 25 documentation findings, 23 were already resolved on `main`: the trademark notice at `README.md:22`, `-p app-cli` at `README.md:102` and `:112`, the live-ledger pointer at `README.md:55`, the historical-gaps label at `README.md:156`, the six-cell `1280×800` row and the reconciled menu-resource evidence in `docs/CS6-PARITY.md`, the panel-status corrections in `docs/NEXT.md`, both optimized builds in `docs/PROGRESS.md`, the `ToolPalette`/`Slices` and `57651` evidence notes in `docs/research/fw-ui-evidence.md`, the remaining-work list in `docs/HANDOFF.md`, the L4 crate row and the right-side-dock description in `docs/wiki/`, the closed menu burn-down in `docs/TASKS.md`, the Aegis baseline re-blessed for decision 0083, and the conflict-free `docs/HANDOFF.md:3` orientation. Two remained valid and are fixed here: `docs/wiki/Parity-Roadmap.md` now states that the 2-inert-of-298 census is not a coverage number and names the 35 `ShellAction::ReportGap` rows, and `README.md:60` labels `docs/GAPS.md` the historical gaps audit instead of current gaps. The per-finding table is in [`docs/tasks/T27-24-pr8-remaining-findings.md`](https://github.com/aliasfoxkde/FireCraft/blob/main/docs/tasks/T27-24-pr8-remaining-findings.md). The same change repairs the Aegis baseline and closes the gap that let it drift: PR #170 merged `docs/tasks/T27-41-windows-arm64-assessment.md` through a documentation-only path, and both CI systems skip the Aegis leg for documentation-only changes — GitHub's `gate` job is gated on `code_changed` while its `docs-only-check` job validates whitespace only, and the GitForge pipeline has no Aegis job at all — so four Windows build literals entered `main` unblessed, and the next code change's gate failed for content it never touched (`main` `c26adffc`: 5 findings the baseline does not cover). The literals are adjudicated as false positives in the `zip-code` class note (`xtask/src/aegis.rs`), the baseline is re-blessed (18 keys added, each with a reason in the table below), and the scan now runs on documentation-only changes in both CI systems — [decision 0132](../decisions/0132-the-secret-scan-runs-on-documentation-changes.md), row T27-62. +**Evidence:** All 9 code findings and 25 documentation findings in PR #8's inline comments were adjudicated against `main` `f1a45bec`. Every code finding is already correct on `main`, so each is rejected with the code that is right: `editor.rs:619` `drag_quadrangle` moves the grabbed corner and the next adjacent corner (`let adjacent = (grabbed + 1) % 4;`), which preserves `TL + BR = TR + BL` and matches its doc comment; `editor.rs:2376` Scale, Skew and Distort read `self.live_selection()` (`editor.rs:2494`, `:2525`, `:2563`, `:2587`), not `document().object_selection`; `editor.rs:2595` Redraw Path sends the sampled pointer trace (`self.redraw_selected_path(&drag.trace)` at `editor.rs:2614`; the signature takes `trace: &[geom::Point]` at `:2733`), guarded by `redraw_path_uses_the_sampled_pointer_trace`; `panels_web.rs:595` the `symbol.createInstance` payload carries only `symbol`, `x`, `y`, `page`, `layer` and `cell`, so `CreateInstancePayload`'s `deny_unknown_fields` is satisfied; `panels_web.rs:574` the placement is `symbol_insert_translation` (`panels_web.rs:275`), which centres the symbol's `definition_content_bounds` on the canvas, guarded by the test at `panels_web.rs:1522`; `retouch.rs:656` the four retouch commands call `brush_diameter` (`retouch.rs:118`), which `checked_mul(2)` and returns `EngineError::BadPayload`; `retouch.rs:1149` `retouch_commands_reject_overflowing_brush_diameters` asserts `Err(EngineError::BadPayload { .. })` for all four; `xtask/src/main.rs:222` `main` carries no `install_pattern_state`, no `temp_dir()` and no `firecraft-aegis-home` — the Aegis leg runs the scanner over the repo root against the content-keyed `.aegis-baseline.tsv` (decision 0074); and the Vector Path node drag that a pure click never ends is fixed on `main` by `79fe2759` (#168), which opens the drag in the press routing (`canvas_ui.rs:1060` and `:1111` → `begin_node_reshape`) and closes it on `drag_stopped` (`canvas_ui.rs:1193` → `:1205`), so no transaction leaks. The one valid code finding, `editor.rs:3271-3276` draft tool ownership, is fixed on `main` by `d3e45cf9` (PR #171) with three regression tests. Of the 25 documentation findings, 23 were already resolved on `main`: the trademark notice at `README.md:22`, `-p app-cli` at `README.md:102` and `:112`, the live-ledger pointer at `README.md:55`, the historical-gaps label at `README.md:156`, the six-cell `1280×800` row and the reconciled menu-resource evidence in `docs/CS6-PARITY.md`, the panel-status corrections in `docs/NEXT.md`, both optimized builds in `docs/PROGRESS.md`, the `ToolPalette`/`Slices` and `57651` evidence notes in `docs/research/fw-ui-evidence.md`, the remaining-work list in `docs/HANDOFF.md`, the L4 crate row and the right-side-dock description in `docs/wiki/`, the closed menu burn-down in `docs/TASKS.md`, the Aegis baseline re-blessed for decision 0083, and the conflict-free `docs/HANDOFF.md:3` orientation. Two remained valid and are fixed here: `docs/wiki/Parity-Roadmap.md` now states that the 2-inert-of-298 census is not a coverage number and names the 35 `ShellAction::ReportGap` rows, and `README.md:60` labels `docs/GAPS.md` the historical gaps audit instead of current gaps. The per-finding table is in [`docs/tasks/T27-24-pr8-remaining-findings.md`](https://github.com/aliasfoxkde/FireCraft/blob/main/docs/tasks/T27-24-pr8-remaining-findings.md). The same change repairs the Aegis baseline and closes the gap that let it drift: PR #170 merged `docs/tasks/T27-41-windows-arm64-assessment.md` through a documentation-only path, and both CI systems skip the Aegis leg for documentation-only changes — GitHub's `gate` job is gated on `code_changed` while its `docs-only-check` job validates whitespace only, and the GitForge pipeline has no Aegis job at all — so four Windows build literals entered `main` unblessed, and the next code change's gate failed for content it never touched (`main` `c26adffc`: 5 findings the baseline does not cover). The literals are adjudicated as false positives in the `zip-code` class note (`xtask/src/aegis.rs`), the baseline is re-blessed (24 keys added — 18 `zip-code`, 4 `ssrf`, 2 `ssrf-localhost` — each with a reason in the table below), and the scan now runs on documentation-only changes in both CI systems — [decision 0132](../decisions/0132-the-secret-scan-runs-on-documentation-changes.md), row T27-62. **Done:** 2026-11-17. **Blockers:** — @@ -90,14 +90,22 @@ The Aegis gate compares the scan against the committed baseline `.aegis-baseline.tsv`, which is content-keyed on `(rule, file, digest)` (decision 0074). PR #170's documentation-only merge added four Windows build literals that no CI job had blessed, so the baseline was missing -18 keys. Each one is a verified false positive, and the class note that -records the reason is in `xtask/src/aegis.rs`: +keys for them. The re-bless adds **24 keys** in total — the four +Windows build identifiers wherever they now appear, the CS6 menu id +`57651`, and the `ssrf` / `ssrf-localhost` keys this change's own new +content introduced (decision 0132, the two installer scripts, the class +note and this ticket). Every one of them is a verified false positive, +and the class note that records the reason is in `xtask/src/aegis.rs`: | Class | Keys added | Why it is a false positive | |---|---|---| -| `zip-code` | 15 — the Windows build identifiers `19041`, `22621`, `26100`, `37710` in `docs/tasks/T27-41-windows-arm64-assessment.md`, `docs/decisions/0132-the-secret-scan-runs-on-documentation-changes.md` and the gate's own class note, plus the CS6 menu id `57651` in `docs/release/EXECUTION-BOARD.md` and this ticket | build identifiers and a menu id, not postal data; the class note names each source | -| `ssrf` | 2 — the node `fetch(url)` fallback in `scripts/install-aegis.sh` and the prose in decision 0132 that names the downloaders | the only URL the repository fetches is the pinned scanner release, whose SHA-256 the installer checks before running it; nothing serves a request on anyone's behalf | -| `ssrf-localhost` | 1 — the loopback fixture server in `scripts/install-aegis-selftest.sh` | a test fixture that serves a local tarball to the installer under test | +| `zip-code` | 18 — the Windows build identifiers `19041`, `22621`, `26100` and `37710` (four keys each in `docs/tasks/T27-41-windows-arm64-assessment.md`, in `docs/decisions/0132-the-secret-scan-runs-on-documentation-changes.md` and in the gate's own class note, `xtask/src/aegis.rs`) and the CS6 menu id `57651` (one key in `docs/release/EXECUTION-BOARD.md`, four in this ticket) | build identifiers and a menu id, not postal data; the class note names each source | +| `ssrf` | 4 — the node `fetch(url)` fallback in `scripts/install-aegis.sh`, the HTTP client dependency the class note names in `xtask/src/aegis.rs`, and the prose in decision 0132 and in this ticket that names the downloaders | the only URL the repository fetches is the pinned scanner release, whose SHA-256 the installer checks before running it; nothing serves a request on anyone's behalf | +| `ssrf-localhost` | 2 — the loopback fixture server in `scripts/install-aegis-selftest.sh`, and the prose in this ticket that names it | a test fixture that serves a local tarball to the installer under test | + +The counts are the baseline's own, not a summary of them: `git show +7bcf28fd -- .aegis-baseline.tsv` adds 24 rows, and `cut -f1` over those +rows counts 18 `zip-code`, 4 `ssrf` and 2 `ssrf-localhost`. The `ssrf` class had no note of its own and fell back to the generic "individually adjudicated" line; it has one now, and the `ssrf-localhost` diff --git a/docs/tasks/T27-62-aegis-scan-runs-on-documentation-changes.md b/docs/tasks/T27-62-aegis-scan-runs-on-documentation-changes.md index e92bbf5..a90d6ea 100644 --- a/docs/tasks/T27-62-aegis-scan-runs-on-documentation-changes.md +++ b/docs/tasks/T27-62-aegis-scan-runs-on-documentation-changes.md @@ -1,6 +1,6 @@ # T27-62 — Run the Aegis secret scan on documentation-only changes, in both CI systems -> **RP:** RP-00 (the release contract's own gates) · **Priority:** P1 · **Status:** IN PROGRESS +> **RP:** RP-00 (the release contract's own gates) · **Priority:** P1 · **Status:** IMPLEMENTED > **Board row:** [EXECUTION-BOARD.md §6](../release/EXECUTION-BOARD.md) > **Depends on:** T27-24 @@ -24,7 +24,7 @@ in both systems; no CI system skips the scan for a documentation-only diff. `scripts/validate_workflow_pins.py`; a documentation-only change on the branch produces a scan in both systems' job logs. -**Evidence:** The defect and its fix are in [decision 0132](../decisions/0132-the-secret-scan-runs-on-documentation-changes.md). The scan covers `docs/`, but both CI systems ran it only for code changes: the GitHub `gate` job is gated on `code_changed` and its `docs-only-check` job validated whitespace only, and the GitForge pipeline had no Aegis job at all while every Cargo job carries a `skip_if` on `CI_CHANGE_INCLUDE_CODE`. PR #170's documentation-only merge therefore entered `main` with four unblessed Windows build literals, and the next code change's gate failed for content it never touched: on `main` `c26adffc` the `gate` job's Aegis leg reported 5 findings the baseline does not cover. The same defect is visible from the other CI system: PR #154 (`8067b53b`, the icon-resource survey) touches no file under `docs/tasks/`, and its GitForge run `fe161636` **succeeded** because the pipeline has no Aegis job to answer, while the GitHub `gate` job on the same SHA **failed** with the same 5 findings, every one of them in `docs/tasks/T27-41-windows-arm64-assessment.md`. The fix has three parts. `scripts/install-aegis.sh` is the single installer — one release URL, one SHA-256, one identity check (`aegis --help` must list the `scan` subcommand, because an unrelated tool also ships under the bare name `aegis`) — and it brings its own fetch, because the authoritative CI image ships no `curl` and no `wget`; the GitHub `gate` job's inline installer now calls it instead of carrying its own copy of the pin. The GitHub `docs-only-check` job installs Rust and the scanner and runs `cargo run --package xtask -- aegis`. The GitForge pipeline gains an `aegis` job with **no `skip_if`**, chained between `release-ledger` and `fmt-layers`, with its own cache directory. The pipeline definition validates (`scripts/validate_gitforge_pipeline.py`: 12 jobs, one root, no join) and the workflow's action pins validate. +**Evidence:** The defect and its fix are in [decision 0132](../decisions/0132-the-secret-scan-runs-on-documentation-changes.md). The scan covers `docs/`, but both CI systems ran it only for code changes: the GitHub `gate` job is gated on `code_changed` and its `docs-only-check` job validated whitespace only, and the GitForge pipeline had no Aegis job at all while every Cargo job carries a `skip_if` on `CI_CHANGE_INCLUDE_CODE`. PR #170's documentation-only merge therefore entered `main` with four unblessed Windows build literals, and the next code change's gate failed for content it never touched: on `main` `c26adffc` the `gate` job's Aegis leg reported 5 findings the baseline does not cover. The same defect is visible from the other CI system: PR #154 (`8067b53b`, the icon-resource survey) touches no file under `docs/tasks/`, and its GitForge run `fe161636` **succeeded** because the pipeline has no Aegis job to answer, while the GitHub `gate` job on the same SHA **failed** with the same 5 findings, every one of them in `docs/tasks/T27-41-windows-arm64-assessment.md`. The fix has three parts. `scripts/install-aegis.sh` is the single installer — one release URL, one SHA-256, one identity check (`aegis --help` must list the `scan` subcommand, because an unrelated tool also ships under the bare name `aegis`) — and it brings its own fetch, because the authoritative CI image ships no `curl` and no `wget`; the GitHub `gate` job's inline installer now calls it instead of carrying its own copy of the pin. The GitHub `docs-only-check` job installs Rust and the scanner and runs `cargo run --package xtask -- aegis`. The GitForge pipeline gains an `aegis` job with **no `skip_if`**, chained between `release-ledger` and `fmt-layers`, with its own cache directory. The pipeline definition validates (`scripts/validate_gitforge_pipeline.py`: 12 jobs, one root, no join) and the workflow's action pins validate. The fix is merged as PR #172 → `7bcf28fd`, and the scan has since run green on a documentation-only change in **both** systems: the GitForge `aegis` job has no `skip_if` and the GitHub `docs-only-check` job installs the scanner and runs it. **Done:** 2026-11-17. **Blockers:** —