Evidence Credit

Resolve <skill-root> to this SKILL.md directory. In Claude Code use ${CLAUDE_SKILL_DIR}; in Codex use the absolute skill path from skill metadata. Resolve <project> to the working project and <trace-dir> to a writable directory outside the installed skill.

1. Run node "<skill-root>/scripts/skill-rails/run.mjs" enter --skill "<skill-root>" unless this context has already processed the returned enter hash.
2. Run node "<skill-root>/scripts/skill-rails/run.mjs" stage --skill "<skill-root>" --project "<project>" --trace-dir "<trace-dir>" --json and save the complete JSON stage result. Supply only requested --judged field=value or --decided field=value values.
3. Stop on a diagnostic, stale snapshot, BLOCK, ASK, or WAIT. Otherwise follow only the current Decision body, stage_artifacts, and ordered effects. Resolve every stage_artifacts[].path from <project>; do not infer replacement paths from collector or authoring files.
4. Attach claims with record --decision <stage-result.json> --type <effect_claimed|proof_recorded|receipt_recorded> --data <json>. For a declared file proof, use --type artifact_verified --artifact <path> --project "<project>". The stage-result file supplies its run id and trace path.
5. Run align --decision <stage-result.json>. Record and alignment evidence is bound to the exact Decision in that file. Missing or agent-only evidence remains unproven; never upgrade it to success.
6. If reinvoke is set, invoke stage again with the same --run-id after effects and save the new stage result. Evidence and alignment for the earlier Decision do not automatically carry to the new Decision.

The runtime decides and verifies; it never performs domain effects. Never edit generated files listed in .generated.json directly.