Parent
Refs #4047. Unit 3 of 9 (sweep unit for the claude-ops plugin, plan Phase 5). Predecessor: #4049 (policy unit), which writes the integration values this unit bakes. Start only after #4049 has merged green.
Agent Brief
Type: Task
Summary: apply the store's integration verdicts for the claude-ops rows to the three audit skills through the audit-native-overlap apply step, per the grammars the policy unit added to the native-references convention. On today's evidence this is a suggest unit, because doctor is registered model-disabled and cannot be composed; if the verdict pass in #4049 chose otherwise for any row, follow the store, not this brief.
Current behavior: audit-install-state carries a route phrase and a Boundary section for doctor; audit-skill-visibility carries a route phrase naming /skill-doctor and /doctor; audit-performance carries nothing for doctor; none of the three declares an unattended argument or records a suggestion under it.
Desired behavior:
audit-install-state: the description phrase stays; the Boundary section's routing paragraph becomes the suggest sentence for /doctor at the end of the run (coverage is partial: doctor fixes, this skill inventories), its basis pointing at the section's existing verification record; the skill declares unattended in its argument-hint and body and records the suggestion under it instead of asking.
audit-skill-visibility: a suggest sentence for /skill-doctor at the start of the run (its coverage of "unused versus cost" is the whole ask when the user wants only that list) and one for /doctor at the end, both with bases pointing at the skill's existing verification record; unattended declared.
audit-performance: no description change and no trim. A suggest sentence for /doctor at the end of the run that opens with a same-plugin pointer to audit-install-state's section for the shared surface facts; unattended declared.
morning-brief is untouched (its row is defer).
- The four store rows take
baked.suggest_sentence: true (or native_step where the store says wrap), the view is regenerated, and every touched skill passes the skill-quality check and stays under the 1,536-character per-entry cap. Version: next minor of claude-ops at PR time, CHANGELOG entry stating the rules.
Key interfaces:
- The three skill bodies and their frontmatter
argument-hint (the unattended argument).
- The store rows for
doctor (three) and skill-doctor (one) and the generated view.
Acceptance criteria:
Out of scope:
Blocked by
Parent
Refs #4047. Unit 3 of 9 (sweep unit for the
claude-opsplugin, plan Phase 5). Predecessor: #4049 (policy unit), which writes theintegrationvalues this unit bakes. Start only after #4049 has merged green.Agent Brief
Type: Task
Summary: apply the store's
integrationverdicts for theclaude-opsrows to the three audit skills through the audit-native-overlap apply step, per the grammars the policy unit added to the native-references convention. On today's evidence this is a suggest unit, becausedoctoris registered model-disabled and cannot be composed; if the verdict pass in #4049 chose otherwise for any row, follow the store, not this brief.Current behavior:
audit-install-statecarries a route phrase and a Boundary section fordoctor;audit-skill-visibilitycarries a route phrase naming/skill-doctorand/doctor;audit-performancecarries nothing fordoctor; none of the three declares anunattendedargument or records a suggestion under it.Desired behavior:
audit-install-state: the description phrase stays; the Boundary section's routing paragraph becomes the suggest sentence for/doctorat the end of the run (coverage is partial: doctor fixes, this skill inventories), its basis pointing at the section's existing verification record; the skill declaresunattendedin its argument-hint and body and records the suggestion under it instead of asking.audit-skill-visibility: a suggest sentence for/skill-doctorat the start of the run (its coverage of "unused versus cost" is the whole ask when the user wants only that list) and one for/doctorat the end, both with bases pointing at the skill's existing verification record;unattendeddeclared.audit-performance: no description change and no trim. A suggest sentence for/doctorat the end of the run that opens with a same-plugin pointer to audit-install-state's section for the shared surface facts;unattendeddeclared.morning-briefis untouched (its row isdefer).baked.suggest_sentence: true(ornative_stepwhere the store sayswrap), the view is regenerated, and every touched skill passes the skill-quality check and stays under the 1,536-character per-entry cap. Version: next minor ofclaude-opsat PR time, CHANGELOG entry stating the rules.Key interfaces:
argument-hint(theunattendedargument).doctor(three) andskill-doctor(one) and the generated view.Acceptance criteria:
bash plugins/skill-quality/scripts/check-skill.sh <skill>passes foraudit-install-state,audit-skill-visibility, andaudit-performance.python3 plugins/claude-ops/skills/audit-native-overlap/scripts/overlap.py self-check --upstream-sha ed404106fcd80ba98ecb7c851e531dcb626d13b7exits 0;grep -c "If /doctor is available in your session (" plugins/claude-ops/skills/audit-install-state/SKILL.mdprints 1;grep -c "unattended" plugins/claude-ops/skills/audit-install-state/SKILL.mdprints at least 2./claude-ops:audit-install-state unattendedyields a report whose final section records the/doctorsuggestion without asking; the transcript excerpt is quoted in the PR body.Out of scope:
claude-opsplugin.integrationvalue; those are the store's and were written in claude-ops/audit-native-overlap: add the integration axis to the store and the wrap and suggest grammars to native-references #4049.Blocked by