ableton-mind-quality-audit - Team Orchestrator

Use this skill for broad quality work: complete audits, command matrix validation, security review, usability and flow review, refactor planning, missing-test coverage, release/runtime readiness and follow-up hardening waves.

Use ableton-mind-build for feature phase execution. Use this skill when the goal is to increase repo quality across existing surfaces.

Team

Phase 0 - Context And Mode

1. Read PLAN.md, _workspace/PROGRESS.md, _workspace/tech-debt.md if present, latest _workspace/qa reports and latest _workspace/quality-audit reports.
2. Read package.json, pyproject.toml, ruff.toml, biome/eslint/tsconfig/vitest config, .github/workflows and Dockerfile.
3. Check git status. Preserve unrelated local changes.
4. Choose mode:

Phase 1 - Briefing

Write _workspace/quality-audit/briefing-{N}.md with:

• User goal and exclusions.
• Current release/project state from _workspace/PROGRESS.md.
• Command matrix sources: package.json, workflows, docs and scripts.
• Side-effect policy for each command class.
• Track assignments and non-overlapping write scopes.
• Gate criteria and expected artifacts.

Command Safety Policy

Classify every command before execution:

In Codex local execution, prefix shell commands with rtk proxy when running raw commands from this repo.

Phase 2 - Parallel Audit

Launch tracks in parallel when their write scopes do not overlap. Read-only tracks may run together. If a track needs to edit files, assign a disjoint ownership boundary and remind it that other agents may be editing the repo.

Required track outputs:

• command-surface-{N}.md: command inventory, run results and stale command findings.
• security-{N}.md: dependency, workflow, bridge and script risk findings.
• usability-{N}.md: setup, doctor, docs, CLI and MCP workflow findings.
• test-coverage-{N}.md: untested surfaces and new tests or proposed tests.
• maintainability-{N}.md: complexity, boundaries, dead code and safe refactor candidates.
• runtime-release-{N}.md: artifact, package, docs, Docker and release-readiness evidence.

Phase 3 - Fan-In Report

The lead writes:

• _workspace/quality-audit/report-{N}.md with findings ordered by severity.
• _workspace/quality-audit/backlog-{N}.md with immediate fixes, later fixes and blocked checks.

Finding format:

• Severity: BLOCKER, MAJOR, MINOR or NICE.
• Owner.
• Evidence: file/line and command or repro step.
• Risk.
• Suggested fix.
• Verification command.
• Status: open, fixed, blocked or accepted debt.

Phase 4 - Fix Wave

When the user asks to fix findings or when the audit scope explicitly includes fixes:

1. Fix BLOCKER and MAJOR findings first.
2. Add or identify tests before behavior changes when feasible.
3. Keep patches scoped to the finding.
4. Re-run focused verification after each fix.
5. Ask qa-integration for cross-boundary verification when TS, Python, recipes, knowledge, docs or packaging contracts changed.
6. Update report/backlog statuses and _workspace/tech-debt.md for deferred debt.

Baseline Verification Matrix

Build the exact matrix from current files, but these commands are expected in this repo when present:

npm run typecheck
npm run lint
npm test
npm run build
npm run build:dxt:check
npm run docs:build
npm run deps:validate
npm run complexity
npm run complexity:py
npm run lint:py
python3 -m unittest discover -s live/AbletonMind/tests -t live -v
npm pack --dry-run --json
npm audit --omit=dev
docker build -t ableton-mind:audit .

Do not pretend hardware, publish, registry or credentialed checks passed. Report them as blocked with exact prerequisites.

Test Scenarios

Normal flow:

• User: "create a team for a complete audit, test every command, check security/usability/refactor/tests."
• Expected: create/read briefing, launch all tracks, classify commands, run safe matrix, write report/backlog, then fix scoped findings only when requested or included in the briefing.

Error flow:

• A command requires publish credentials, Ableton Live, Push hardware or external registry write access.
• Expected: do not run it by default; record it as blocked with the exact command, required prerequisite and safer dry-run alternative.

Follow-up flow:

• User: "continua a auditoria" or "corrige os achados."
• Expected: read latest _workspace/quality-audit report/backlog, avoid rediscovery, patch bounded items with tests and update statuses.