Repository navigation
feat(k9): vnd.k9.coordination subtype, coordination grammar v1.2.0, native signed-push - #1233
Merged
Merged
Conversation
Owner ruling, 2026-10-13: the k9-coordination dialect gets its own subtype rather than a parameter of application/vnd.k9. File-type detectors drop parameters, which would label coordination files as Nickel components; a subtype survives that and lets each type state its own encoding and security considerations. - New registration draft for application/vnd.k9.coordination beside the coordination spec (pure data, YAML-like body under its own ABNF, .k9). - vnd.k9 registration (Revision 4) and its form-ready copy: the open decision is settled, the type claims only .k9.ncl, and it is ready to file together with the companion type. - SPEC.adoc MIME recipes: separate freedesktop entries (globs decide; a nested "metadata:" match tells extensionless files apart, verified against every committed .k9 and .k9.ncl file), a UTI for the new type, and a mime.types note that it cannot express .k9.ncl. - K9-CONTRACT-SPEC and COORDINATION-K9-SPEC name the media type of each dialect. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GqfpQtLuMczxtgG1U9PkLW
Owner ruling, 2026-10-13: extend the grammar for what belongs in a coordination file, and move the rest out. Grammar v1.2.0 (coordination-k9-grammar_v1.2.abnf), additive: every v1.1.1 file is a v1.2.0 file. - build_commands (between project and invariants): command + purpose. - innervation (between architecture and do_not_create): afferent signals with their consumers, efferent signals with auto_fix, and reflex guards. - Comments are ASCII as RFC 5234 requires (bar the SPDX copyright line). The v1.1 file stays, marked superseded, so existing links resolve; the spec, READMEs, CLAUDE.md, AGENTS/REGISTRY and the K9 contract now point at v1.2. Root coordination.k9 now parses under v1.2.0 (checked with an ABNF parser; it fails v1.1, as it should). Previously it failed at its `---` line. - `---` removed; blank lines inside sections removed; repo and glob values unquoted; multi-word terminology and ecosystem names made tokens; build_commands restructured; innervation consumers are lists. - Moved out: ports / project.verisimdb_port to ECOSYSTEM.a2ml [ports], project.crg_grade to STATE.a2ml [position]. - Stale facts corrected: languages now match the tree (AffineScript and Julia in; ReScript, Agda, Elixir out; none are present); protected entries for the deleted a2ml/ and lol/ dropped; the banned-language rule and the *.ts reason follow the current language policy. - NOT changed: the pmpl-license-only invariant contradicts this repo's MPL-2.0 classification, but licence content is flag-only. The generator's parser reads the new file (consumers now parse as a list instead of a string). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GqfpQtLuMczxtgG1U9PkLW
…ed.sh The signed-push composite action delegated its push to Asana/push-signed-commits, which is Python (banned here), has no licence file at the pinned commit, installs Python through the tag-pinned actions/setup-python@v2 and runs an unpinned `pip install requests` at run time. The same work is a short bash script over `gh api graphql` createCommitOnBranch, the pattern scripts/regen-dependabot-locks.sh already uses. push-verified.sh keeps the contract (remote branch must exist and must not have diverged; local commits replayed oldest first with their messages; each call names the head it expects, so a concurrent push fails closed; local branch reset to the remote afterwards) and is stricter: - file contents are base64'd from the raw blob, so binary files survive (the Python action decoded them as UTF-8 text); - merges, symlinks, submodules, executable additions and mode changes, none of which createCommitOnBranch can express, are refused before anything is sent, and the pushed tree is compared with the local one. scripts/tests/signed-push-test.sh drives it against a local bare remote through a stub gh that applies each mutation as GitHub would: 13 cases, including a partial-push report and the five refusals. Writing it caught a merge being detected only after earlier commits had been pushed; every commit is now validated and prepared before the first call. A mutant that stops sending deletions fails 4 of the cases. The asana/push-signed-commits and actions/setup-python@v2 records stay in actions.lock for now: nothing references them, every lock gate passes with them present, and the file is regenerated by tooling, not by hand. The next regeneration drops them. The regen job's composite guard (#1122) is unchanged. prune-stale still cannot see refs a workflow reaches through a local action, so this repo still needs a hand-run regeneration after action bumps. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GqfpQtLuMczxtgG1U9PkLW
Contributor
K9 contract conformancerun https://github.com/hyperpolymath/standards/actions/runs/38475956439 K9 normative contract typecheckK9 contract self-testK9 conformance fixturesK9 corpus conformance (L2) |
Contributor
|
Note Currently processing new changes in this PR. This may take a few minutes, please wait... ⚙️ Run configuration
📒 Files selected for processing: 24
✨ Finishing Touches: 💡 1
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
hyperpolymath
marked this pull request as ready for review
October 13, 2026 22:12
Clears the shellcheck warnings CodeFactor raised on the new suite: SC2164 (cd without a fallback inside the fixture subshells) and SC2181 (the second-run assertion read $? instead of testing run directly). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GqfpQtLuMczxtgG1U9PkLW
Contributor
K9 contract conformancerun https://github.com/hyperpolymath/standards/actions/runs/38476322882 K9 normative contract typecheckK9 contract self-testK9 conformance fixturesK9 corpus conformance (L2) |
A2ML is retired (CLAUDE.md, Machine-Readable Artefacts: never add a fact to a .a2ml file), so the two facts moved out of coordination.k9 in 1cc0fc0 should not have gone into STATE.a2ml and ECOSYSTEM.a2ml. Both files are back to their main content. - The VeriSimDB port already has an authoritative home: .verisimdb/config.toml (port = 8097). The ECOSYSTEM.a2ml copy was a duplicate. - The CRG grade (B, self-declared) belongs in a root READINESS.md per component-readiness-grades/COMPONENT-READINESS-GRADES.adoc, which `just crg-grade` reads. The repo has no READINESS.md, so the header comment in coordination.k9 records the grade and where it belongs. coordination.k9 still parses under grammar v1.2.0; descriptile-policy and validate-k9 pass. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GqfpQtLuMczxtgG1U9PkLW
Contributor
K9 contract conformancerun https://github.com/hyperpolymath/standards/actions/runs/38476549165 K9 normative contract typecheckK9 contract self-testK9 conformance fixturesK9 corpus conformance (L2) |
|
Contributor
|
❌ Failed to create Coding Agent finishing-touch task. Please try again. |
Contributor
|
Add Carrot credits or activate Agent usage billing to use Autofix |
hyperpolymath
pushed a commit
that referenced
this pull request
Oct 13, 2026
The README is prose documentation, which LICENCE-POLICY Rule 1 puts under CC-BY-SA-4.0. It was given MPL-2.0 in error when it was added in #1233. Owner-approved change to this one file (2026-10-13). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GqfpQtLuMczxtgG1U9PkLW
hyperpolymath
added a commit
that referenced
this pull request
Oct 13, 2026
…e; signed-push README licence (#1235) Follow-up to #1233. The repo that defines CRG had no machine-readable grade declaration, so `just crg-grade` would have printed X for it. In fact the recipe never ran at all; see below. ## `READINESS.md` (new) - Holds the `**Current Grade:** C` line that the CRG spec asks for (COMPONENT-READINESS-GRADES.adoc, "Machine-Readable Grade Declaration"). - **Why C (owner ruling 2026-10-13):** C is the latest assessment, `0-canon/CRG-SELF-ASSESSMENT.adoc` (2026-04-09). That document also lists the gaps to Grade B. - `component-readiness-grades/SELF-ASSESSMENT.adoc` records an earlier B (2026-04-04). `READINESS.md` notes it and says the later assessment governs. That document is not edited here. - Licensed CC-BY-SA-4.0 as prose documentation, per LICENCE-POLICY Rule 1. ## `just crg-grade` / `just crg-badge` fixed - **The bug:** both recipes used Make's `$$` escape, which `just` does not have. `sh` therefore received `grade=<pid>(grep …)` and failed with `Syntax error: "(" unexpected`. - **The fix:** the seven `$$` in those two recipes become `$`. Nothing else in the Justfile changes. - **Checked with just 1.21:** - `crg-grade` prints `C`; - `crg-badge` prints the yellow CRG C shields.io badge; - without `READINESS.md`, both fall back to `X`. ## `signed-push` README licence (owner-approved) `.github/actions/signed-push/README.md` was added in #1233 with an MPL-2.0 header. It is prose documentation, so LICENCE-POLICY Rule 1 puts it under CC-BY-SA-4.0. The owner approved changing this one file. Only the SPDX line changes. ## Not changed (flagged) - The B assessment in `component-readiness-grades/SELF-ASSESSMENT.adoc` stays as it is. - So does the header comment in `coordination.k9`, which still says the grade is B and that this repo has no `READINESS.md` yet. It is now stale. - The CRG spec says these recipes come from `rsr-template-repo`'s Justfile. If that file has the same `$$`, it has the same bug. It is a different repository, so it is not fixed here. ## Testing - `just --list` parses the Justfile, and the two recipes were run as described above. - Shell suite: 77 of 81 test files pass. The other 4 fail identically on `main` in this sandbox, which lacks the tools they need: KYAML-capable yq, the JCS tool and Ruby. The 4 are `check-actions-lock-gate`, `check-ijson-jcs`, `kyaml-format` and `science-ci-security`. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01GqfpQtLuMczxtgG1U9PkLW --- _Generated by [Claude Code](https://claude.ai/code/session_01GqfpQtLuMczxtgG1U9PkLW)_ --------- Co-authored-by: Claude <noreply@anthropic.com>
hyperpolymath
added a commit
that referenced
this pull request
Oct 13, 2026
…1238) ## Summary [#1233](#1233) already did the first half of this job. In `.github/actions/signed-push/action.yml`, the "Push local commits as Verified" step now runs `push-verified.sh`, a bash script beside the action. It uses `gh` and `jq`, the App installation token from `steps.app-token`, and `createCommitOnBranch` with `expectedHeadOid`, so if the remote head moves the push fails closed. The inputs did not change: `app-id`, `private-key`, `branch`. The action has no outputs. It never had any. #1233 left the lock records in place on purpose. This PR removes them. Nothing references either one any more: | Record removed from `dependencies:` | Why it was safe to remove | |---|---| | `asana/push-signed-commits@d615ca88…` (ref `v1.3`) | No `uses:` anywhere names it, and neither does any `workflows:` list. | | `actions/setup-python@v2` | Its only reference was the Asana record's nested `uses:`. It was also the only key in the lock that is not a SHA. | The diff deletes 12 lines and changes nothing else. ### How the records were removed The repo's lock tooling cannot do this: - `prune-stale.sh` only edits the `workflows:` section. - `close-lock.sh` only adds records. - `gh actions-lock` regeneration is fenced off for this repo because of the composite action (#1122). So I removed the two `dependencies:` blocks with a gawk filter that only touches that section. Afterwards `prune-stale.sh` was a no-op, and `close-lock.sh` found 0 dangling refs and left the file alone. ## Testing All lock gates were run before and after the deletion, with the pinned tool versions: | Gate | Result | |---|---| | `scripts/check-lock-sync.sh` | in sync | | `.githooks/validate-actions-lock.sh` | 25 refs found among 27 lock keys (was 28) | | `scripts/check-actions-lock-gate.sh` (gh-actions-lock v0.1.6, the version CI pins) | "Immutable direct and transitive lockfile coverage verified" | | `scripts/prune-stale.sh` | no-op | | `scripts/close-lock.sh` | 0 dangling refs | | `scripts/lock-selfcheck.sh HEAD` | SELF-CONSISTENT | Script tests that touch this action or the lock all pass: | Test | Result | |---|---| | `validate-sha-pins-test` | 22/22 | | `signed-push-test` | 13/13 | | `check-actions-lock-gate-test` | 23/23 | | `validate-actions-lock-test` | 5/5 | | `check-lock-sync-test` | 10/10 | | `prune-stale-test` | 15/15 | | `close-lock-test` | 9/9 | | `lock-selfcheck-test` | 5/5 | | `regen-dependabot-locks-test` | 34/34 | | `update-actions-lock-test` | 15/15 | | `kyaml-format-test` | 17/17 | The full `run-shell-test-suite.sh` passed 79 of 81 files. The other two fail the same way on `main` in this container because tools are missing: - `check-ijson-jcs-test` needs the `ijson-jcs` binary. - `science-ci-security-test` needs `ssh-keygen`. Neither test reads `actions.lock`. The pre-commit hook passed in full (gitleaks included), and the commit-msg hook passed. `signed-push-smoke.yml` was **not** run. It is dispatch-only and needs `APP_ID`/`APP_PRIVATE_KEY`, which `config/README.adoc` says this repo does not hold yet. ## Follow-up, not done here (outside this PR's scope) - **`.hypatia-baseline.json`** has an `invalid_actions_lock` acknowledgement (`**actions.lock`, tracking #963, expires 2026-12-22) for exactly this defect. Its exit condition was "we replace the action", and that is now met. Removing the entry would shrink the ledger, which the ratchet allows. I left it because I could not run Hypatia here to confirm that nothing else in the lock triggers the rule. - **The composite guard in `regen-dependabot-locks.sh` (#1122)** is unchanged. The action's only third-party dependency is now `actions/create-github-app-token`, and other workflows also use it directly. So a regeneration should no longer have to restore records by hand. Relaxing the guard still needs the live check #1233 described. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_014h2gFLLQibfB5N5hFz9ZCD --- _Generated by [Claude Code](https://claude.ai/code/session_014h2gFLLQibfB5N5hFz9ZCD)_ Co-authored-by: Claude <noreply@anthropic.com>
This was referenced Oct 14, 2026
hyperpolymath
added a commit
that referenced
this pull request
Oct 14, 2026
Closes #963. ## Summary Removes the `.hypatia-baseline.json` acknowledgement for `invalid_actions_lock`. The entry had: - `rule_module: ["workflow_audit", "workflow_hardening"]` - `file_pattern: "**actions.lock"` - tracking issue #963 - `expires_at: 2026-12-22` It acknowledged `asana/push-signed-commits@d615…`, which declared the floating transitive `actions/setup-python@v2`. One of its listed exit criteria was "we replace the action", and that is now met: - #1233 replaced the Asana action in `.github/actions/signed-push/` with `push-verified.sh`. - #1238 deleted both the `asana/push-signed-commits` and the `actions/setup-python@v2` records from `actions.lock`. While the entry stood, `**actions.lock` would have muted **any** new `invalid_actions_lock` finding in the repo until 2026-12-22. ## Evidence the rule no longer fires on `main` These are from the Hypatia scan on `main` at `7d2041a` ([run 38484072239](https://github.com/hyperpolymath/standards/actions/runs/38484072239), scanner pinned at `51ab6496`): - The scan reported 59 findings: critical=2, high=2, medium=34, warn=21. - `findings_kept` is `[]`. - None of the 59 suppressed findings has type `invalid_actions_lock`. The two highs are WH013 and `ncl_k9_missing_pedigree`. - No finding names `actions.lock`, asana or setup-python anywhere in the log. The ack was matching nothing. Both lock checks from the task print nothing on this branch: - the non-SHA key grep; - the non-SHA nested `uses:` grep. ## Changes | File | Change | |---|---| | `.hypatia-baseline.json` | One entry deleted, 12 lines. jq confirms the result equals the original with only that entry removed. No other entry changes. | | `scripts/tests/apply-baseline-test.sh` | The "SHIPPED ledger" section asserted that the real baseline holds **exactly one** such ack and suppresses both emission paths, so it would go red on this removal. It is inverted: it now asserts the ledger holds **no** `invalid_actions_lock` ack and that both emission paths **reach the gate** (`workflow_audit`/`actions.lock` and `workflow_hardening`/`.github/workflows/actions.lock`). The fixture-based tests for list-valued `rule_module` are unchanged. | `docs/HYPATIA-BASELINE-FORMAT.adoc`, the schema description and the comment in `apply-baseline.sh` still use `invalid_actions_lock` as their example of one defect emitted by two modules. That describes how the scanner behaves, not a live ack, so they are left as they are. ## Testing | Check | Result | |---|---| | `bash scripts/tests/exemption-ratchet-test.sh` | 14 passed, 0 failed | | `bash scripts/tests/apply-baseline-test.sh` | 32 passed, 0 failed | | `bash scripts/tests/hypatia-blocking-gate-test.sh` | exit 0 | | `scripts/check-exemption-ratchet.sh origin/main` | `SHRANK .hypatia-baseline.json: 189 -> 188`. OK, and no `Ratchet-exception:` was needed. | | `apply-baseline.sh` on `[]` in blocking mode | schema-valid, exit 0 | | `apply-baseline.sh` with both `invalid_actions_lock` emissions in blocking mode | kept=2, suppressed=0, **exit 1**. A recurrence now blocks. | | `.githooks/pre-commit` and `commit-msg` | passed. Included gitleaks 8.28.0, validate-actions-lock and validate-sha-pins, using checksum-verified gitleaks and mikefarah yq v4.47.1. | The Hypatia scanner itself was **not** run locally, because the container has no Elixir. This PR's own `scan / Hypatia Neurosymbolic Analysis` run is the live confirmation. Licence headers are untouched. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01TbbnKqskQA5mj5gfMniPzi Co-authored-by: Claude <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Follow-up to #1232. This PR covers its three open items, each in its own commit, plus two fix-ups.
1.
application/vnd.k9.coordination(owner ruling 2026-10-13: own subtype, not a parameter)2-protocols/k9-coordination/docs/iana/application-vnd.k9.coordination-registration.txt. The format is pure data, with a YAML-like body governed by its own ABNF, and uses the.k9extension.vnd.k9registration (Revision 4) and its form-ready copy:.k9.ncl;SPEC.adoc:metadata:match tells the two apart. Checked against every committed.k9and.k9.nclfile.mime.typescannot express.k9.ncl.2. Coordination grammar v1.2.0; the repo's own
coordination.k9conforms(Owner ruling: extend the grammar for what belongs, move the rest out.)
coordination-k9-grammar_v1.2.abnfadds two optional sections:build_commandsandinnervation(afferent / efferent / reflex). The change is additive, so every v1.1.1 file is still valid.coordination.k9now parses under v1.2.0. I checked this with an ABNF parser (apg-js), and the file correctly fails v1.1.---line.ports,verisimdb_portandcrg_grade. A2ML is retired, so neither fact went into a.a2mlfile:.verisimdb/config.toml.READINESS.mdunder the CRG spec, andjust crg-gradereads it from there. This repo has noREADINESS.mdyet, so thecoordination.k9header comment records the grade and where it belongs.a2ml/andlol/are gone;pmpl-license-onlyinvariant contradicts this repo's MPL-2.0 classification. Licence content is flag-only.consumersnow parses as a list instead of a string.3.
signed-pushwithoutAsana/push-signed-commitsThe Asana action is written in Python and has no licence file. It also uses the tag-pinned
setup-python@v2and runs an unpinnedpip install. It is replaced by.github/actions/signed-push/push-verified.sh, which uses bash,ghandjqto callcreateCommitOnBranch.scripts/tests/signed-push-test.shruns 13 cases against a stub GitHub, with no shellcheck warnings. Writing it caught a real partial-push bug in my first draft, which is now fixed. A mutant that stops sending deletions fails 4 cases.asana/push-signed-commitsandactions/setup-python@v2records are not hand-edited out ofactions.lock. All lock gates pass with them present, and the next tool regeneration drops them.prune-stalestill cannot see refs that a workflow reaches through a local action, so action bumps here still need a hand-run regeneration. Relaxing the guard safely needs a live check of how GitHub validates locks for local composites.Testing
minimal.k9, the spec's new examples and the rootcoordination.k9pass v1.2. Negative controls fail.k9-validate.sh --self-testand thevalidate-k9hook pass.signed-push-test13/13,validate-sha-pins-test22/22 (with mikefarah yq),descriptile-policy-test11/11. The docstring scan shows 100% on the new code.validate-actions-lockandcheck-lock-sync. These also pass:check-workflow-duplicate-keys,check-standards-map,check-canon-lockstep.mainhere, because this sandbox lacks KYAML-capable yq, the JCS tool and Ruby. CI is the authority for those four:check-actions-lock-gate,check-ijson-jcs,kyaml-formatandscience-ci-security.🤖 Generated with Claude Code
https://claude.ai/code/session_01GqfpQtLuMczxtgG1U9PkLW
Generated by Claude Code