Repository navigation
chore(state): archive STATE.a2ml and rewire the gates that required it - #78
Conversation
Completes state-v1-decision option B (ruled 2026-11-10). The thin journal
(docs/status/JOURNAL.adoc) and the repo deed now carry what STATE v1 carried, so
STATE.a2ml is renamed to .machine_readable/archive/state-v1.a2ml with a tombstone.
All 8 sections are kept verbatim, so no milestone row or blocker is lost.
Gates that failed without the file now require its successor, one for one:
- openssf-compliance.yml: requires docs/status/JOURNAL.adoc, and the
placeholder-token scan covers it
- the Must invariant six-directory-present, validate-template.sh, and the e2e
template test
Each was checked both ways: it passes with the journal and fails loudly without it.
Readers of STATE content moved to the new homes:
- arrival-pack: lifecycle and maturity come from the deed (status, maturity);
the derivable completion figure is dropped; provenance hashes the deed.
CLAUDE.md is regenerated with the real generator (nickel 1.18) and
verify.sh passes. The estate-common STATE row and read order now name the
journal, a deliberate divergence from the draft estate block.
- coaptation extractor: status and maturity from the deed, phase from the
journal. completion-percentage, open-warnings, open-failures and last-result
have no source any more and are emitted EMPTY, so they register as ABSENT and
never as a recorded zero.
- Justfile and contractiles/Justfile: info, state-touch, state-phase.
- the k9 methodology guard, AI-tool rules, 0.1-AI-MANIFEST, the descriptiles
README and 0-AI-MANIFEST invariant (it said 7 core files MUST exist), the PR
template, AI-CONVENTIONS, both deed headers, and the ROADMAP headers.
Baseline comparison on unmodified main, same commands both sides:
- validate-template.sh output is byte-identical; the e2e test stops at the same
pre-existing step; the A2ML validator reports the same file counts.
- systemet's CLAUDE.md drift gate was already red on main (STATE changed after
CLAUDE.md was generated, and it claimed "phase design; 15% complete"); the
regeneration clears that.
- the coaptation receipt gate is red on main in both repos (the committed
receipt is stale: repo = "rsr-template-repo"). It is NOT regenerated here:
standards says never to regenerate an A2ML file.
Left alone on purpose: the canon-derived hook comments in AGENTIC/PLAYBOOK, the
"never create STATE.a2ml" rules, the template example scripts, the historical ADR,
and the generic RSR/CRG guidance in docs/. They name STATE as a family, not as a
path anything reads.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J1XmMC49T8sZ3uUuvrgB7y
Finishing Touches
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. |
…eader SonarCloud flagged 6 MAJOR code smells on the lines this PR added, in extract.sh and extract-facts.sh: a single-bracket test where [[ is safer (S7688), and positional parameters used directly in deed_sym (S7679). Use [[ -f ... ]] in the two new existence checks, and name the clause and keyword parameters in deed_sym. Pre-existing lines in the same files are left as they were; the scanner only reports new code, and rewriting them would widen the PR. No behaviour change: the extractors emit the same lifecycle, maturity and fact presence, verify.sh still passes, and a missing deed still aborts with exit 2. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01J1XmMC49T8sZ3uUuvrgB7y
|
This PR edits three No fix exists yet. That content debt, and the K9 spec question about the SonarCloud: the quality gate passes, but it reported 6 new MAJOR code smells, all in shell I added (
Generated by Claude Code |
|
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
Summary
Completes STATE option B (ruled 2026-11-10). The thin journal and the repo deed now carry what STATE v1 carried, so
STATE.a2mlis renamed to.machine_readable/archive/state-v1.a2mlwith a tombstone.gitrecords it as a rename. All 8 sections are kept verbatim, so no milestone row or blocker is lost.Changes
Gates that failed without the file now require its successor, one for one
openssf-compliance.ymlrequiresdocs/status/JOURNAL.adoc, and its placeholder-token scan covers it.six-directory-present,scripts/validate-template.sh, and the e2e template test.Readers of STATE content moved to the new homes
(status)and(maturity). The derivable completion figure is dropped. Provenance now hashes the deed.CLAUDE.mdis regenerated with the real generator.completion-percentage,open-warnings,open-failuresandlast-resulthave no source any more. They are emitted empty, so they register as absent and never as a recorded zero.Justfileandcontractiles/Justfile(info,state-touch,state-phase), the k9 methodology guard, AI-tool rules,0.1-AI-MANIFEST, the6a2README, the PR template,AI-CONVENTIONS, the deed header, and the ROADMAP/PROOF-STATUS pointers.6a2/0-AI-MANIFEST.a2mlsaid "the 7 core A2ML files MUST exist". That is now false, so it says 6 and points at the archive.How it was checked
I installed
nickel1.18 andjust1.59, then ran each rewired gate against unmodifiedmainand against this branch.verify.sh, realnickel)main; passes herevalidate-template.shmain(its failure, a missingts-blocker.yml, is pre-existing)METADATA_FILESloop I edited passes when run directlyvalidate-template.sh, the Must invariant, the openssf logic and the coaptation extractor each fail loudly; the same Must command passes with the journal presentjust info/state-phase/state-touchstate-touchedits exactly the:updated:line and theupdated::valuedeed_lint.js,check-root-shape,check-no-md-in-docsgrepfor live code still namingSTATE.a2mlDecisions for you
CLAUDE.mdnow diverges from the draft estate block. Its read order and file table name the journal instead of STATE, and the State line shows lifecycle and maturity only. The same change belongs in the estate template upstream (standards/rsr-template-repo), or other repos will regenerate the old text.just validate-coapt) is red onmainand stays red. Its committed receipt is stale (repo = "rsr-template-repo"). I did not regenerate it, because standards says never to regenerate an A2ML file. Regenerate it, or retire the gate?6a2/AGENTIC.a2mlandPLAYBOOK.a2mlstill say "Read … STATE.a2ml". Editing them would make them diverge from template canon. They need fixing in the template.Left alone on purpose
The "never create
STATE.a2ml" rules, theself-validatingtemplate example scripts, the historical ADR (docs/decisions/0001), and the generic RSR/CRG guidance indocs/(READINESS, CRG-AUDIT-TEMPLATE, RSR_OUTLINE, and others). They name STATE as a family, not as a path anything reads. The sections some of them cite ([dogfooding-status]and so on) never existed in this repo's STATE.RSR Quality Checklist
Required
mainand this branch.deed_lint.jsOK.unsafe; no banned functions..envfiles included.As Applicable
docs/status/JOURNAL.adocupdated: the resolvedstate-archive-rewireblocker is removed.Generated by Claude Code