| name | arch-pr-analyzer |
| description | Analyze PR or CI architecture quality from ArchSig measurement packets, AIR, theorem-check, feature-report, policy, signature diff, and pr-quality-analysis artifacts. Use when Codex is asked to review a PR's architecture impact, inspect CI architecture reports, explain current architecture state for a change, or recommend bounded PR review actions. |
Arch PR Analyzer
Purpose
Analyze PR / CI architecture evidence and produce review cues. This skill simplifies ArchSig's PR surface for Codex: it may run the necessary fieldsig commands, read generated artifacts, and explain architecture risks without treating tool output as merge approval, Lean proof, global architecture truth, or causal diagnosis.
Inputs To Prefer
signature-diff-report/v0.5.0
- Sig0 and validation reports
archsig-measurement-packet/v0.5.4 and its canonical ArchSig input-validation evidence
- AIR, AIR validation, theorem-check, feature-report
aat-observable-bundle/v0.5.0 and validation report when the PR review asks for AAT concept coverage
architecture-policy/v0.5.0 and law-violation-report/v0.5.0
- policy-decision and PR comment summary
pr-quality-analysis-report/v0.5.0
sft-review-summary/v0.5.0 only when the user asks for SFT review judgement or future-impact review
Do not use planning forecast artifacts for PR merge review unless the user explicitly asks for planning context.
Minimal Execution Rules
- Use
${FIELDSIG_BIN:-fieldsig} by default.
- Keep generated files under
.archsig/ unless the user specifies another directory.
- If
${FIELDSIG_BIN:-fieldsig} --help fails, ask for FIELDSIG_BIN=/path/to/archsig.
- Stop downstream commands after an upstream failure.
- Do not require the ArchSig source repository.
Workflow
-
Establish the PR evidence boundary.
- Identify before / after revisions, supplied artifacts, missing artifacts, private refs, unmeasured axes, and non-conclusions.
- Read
references/pr-analysis-playbook.md for non-trivial PR review.
- Read
references/artifact-reading-guide.md when artifact meaning is unclear.
-
Generate missing PR artifacts when appropriate.
${FIELDSIG_BIN:-fieldsig} --root . --out .archsig/signature/current/sig0.json
${FIELDSIG_BIN:-fieldsig} validate \
--input .archsig/signature/current/sig0.json \
--out .archsig/signature/current/validation.json \
--universe-mode local-only
If before / after snapshots are available, use signature-diff. If an ArchSig measurement packet is available, project it into the FieldSig handoff:
${FIELDSIG_BIN:-fieldsig} archsig-analysis-sft-input \
--measurement-packet .archsig/archsig/measurement-packet.json \
--out .archsig/sft/operation-support-estimate.json
If a project-local law policy is supplied, validate it and evaluate Layered Architecture findings:
${FIELDSIG_BIN:-fieldsig} architecture-policy \
--input .archsig/policy/architecture-policy.json \
--out .archsig/policy/architecture-policy-validation.json
${FIELDSIG_BIN:-fieldsig} law-violation-report \
--sig0 .archsig/signature/current/sig0.json \
--policy .archsig/policy/architecture-policy.json \
--out .archsig/policy/law-violation-report.json
Validate supplied PR quality report when present:
${FIELDSIG_BIN:-fieldsig} pr-quality-analysis \
--input .archsig/pr/pr-quality-analysis-report.json \
--out .archsig/pr/pr-quality-analysis-validation.json
Validate supplied AAT observable bundle when the review asks for invariant / witness / operation /
theorem-boundary coverage:
${FIELDSIG_BIN:-fieldsig} aat-observable-bundle \
--input .archsig/aat/aat-observable-bundle.json \
--out .archsig/aat/aat-observable-bundle-validation.json
If an SFT review judgement artifact is supplied, validate it separately from deterministic PR quality analysis:
${FIELDSIG_BIN:-fieldsig} sft-review-summary \
--input .archsig/sft/sft-review-summary.json \
--out .archsig/sft/sft-review-summary-validation.json
- Produce the review.
- Separate measured zero from unmeasured.
- Separate warning, failure, missing evidence, and negative evidence.
- Treat Layered Architecture
deterministicViolations[] as deterministic over resolved selectors and measured edges.
- Treat SRP
semanticRole / responsibilityRegions / reasonToChange / lawRefs as evidence for LLM judgment, not as tool-only violation.
- For SRP probable violation, cite both evidence refs and policy refs and distinguish
acceptableOrchestrator, allowedException, and unmeasured.
- Tie each finding to artifact ids or paths.
- When using
aat-observable-bundle/v0.5.0, translate reviewActions into evidence gap,
blocked formal claim, review guardrail, or next evidence without treating LLM judgment as proof.
- When using
sft-review-summary/v0.5.0, report opened futures, closed futures, boundary failures,
and next actions with evidence refs. Do not treat the summary status as merge approval.
- Keep recommendations bounded to PR review actions.
Output Shape
PR architecture state
- ...
Review cues
- ...
Evidence gaps
- ...
Recommended next actions
- ...
SFT review judgement
- opened futures:
- closed futures:
- boundary failures:
- next actions:
Non-conclusions
- ...
Use Japanese for user-facing reports in this repository, unless the user asks otherwise.