Skip to content

qms-skill-management

Consolidated skill-corpus QMS toolkit: one skill, three routes covering spec review, promotion certification, and corpus ops.

  • Area: Foundational
  • Category: utilities
  • Level: advanced
  • Input: conversation
  • Output: document
  • Setup: none
  • Roles:

Runs skill QMS reviews, certification, and corpus operations.

How you ask

  • “Spec-review the humanize-content skill”
  • “Certify this skill for release promotion”
  • “Run a full skill QMS audit”
  • “Rerun spec reviews for all governed skills”
  • “Remediate the weakest skills”
  • “Generate a current-state corpus report”

Inputs

  • skill_path — Skill folder name, passed to batch_spec_review.py as --filter (omit for batch mode)
  • batch — If true, review all skills under --skills-dir
  • skill_path — Path to the skill directory
  • cert_threshold — Minimum spec-review score the gate accepts, default 8.0. Raising it blocks skills the standard bar would approve; lowering it is how a deliberate exception is made visible rather than argued in prose.
  • mode — One of integrity-check | scorecard-rerun | spec-review-rerun | remediation | auto-optimize | tessl-review | corpus-report

Outputs

  • Approve-or-block verdict with every failing check named. certify_skill.py has no write path.
  • Per-mode report or log; actual artifact path depends on --mode (see Route 3 sub-sections).
  • Written by the agent for mode remediation — per-skill before/after spec score, delta, KEEP/REVERT result and the fix applied, then the tranche summary.
  • Written for mode auto-optimize by skillsmith's run_skill_loop.py — per-skill eval-loop outcome (kept or reverted) for the skills carrying an eval manifest, and the stop reason.
  • spec-review-json (json)
  • {company.qms.audits}/optimization-log-{YYYY-MM-DD}.md
  • {company.qms.audits}/remediation-log-{YYYY-MM-DD}.md
  • {company.qms.spec-reviews-individual}/{skill}-{YYYY-MM-DD}.json