| name | archsig-pr-reviewer |
| description | Compare ArchSig base/head measurement runs, apply gate policy, read the changed code, and turn the result into bounded source-first review questions. |
ArchSig PR Reviewer
Use this skill when asked to run ArchSig-assisted PR review.
Inputs
- Base run directory containing
archsig-measurement-packet.json.
- Head run directory containing
archsig-measurement-packet.json.
- Gate policy JSON.
- Source diff or PR checkout for human review.
- Prefer an
archsig-policy-bundle/v0.5.4 for both base and head runs when
the policy, law surface, and profile are intended to remain fixed.
Workflow
- Run
archsig analyze for the base and head contexts when run directories are not already supplied; use --policy-bundle when the bundle is supplied and confirm the three component fingerprints.
- Run
archsig compare --base-run <base> --head-run <head> --out-dir <compare>; the command derives the selected coarse-to-fine relation from the two normalized ArchMap runs.
- Run
archsig gate --packet <head>/archsig-measurement-packet.json --policy <policy> --comparison <compare>/archsig-comparison-report.json --out <gate-report>.
- Read the comparison report, gate report, summary, insight report, and changed source.
- Write bounded review questions. Do not treat ArchSig output as merge approval, forecast truth, or source-code proof.
Output
Report:
- gate decision and blocking rules;
- introduced, cleared, and preexisting measurement transitions;
- source evidence checked for each high-priority finding;
- concrete follow-up questions or patch suggestions.
For SAGA packets, review lawDependent and lawIndependent separately. A
class-zero finding must cite the derived normalized-ArchMap context relation;
the compare report records the named reason when that relation is unavailable.