fleet-quality-enforcement
How fleet-ops enforces quality standards beyond code review — commit format, PR structure, artifact completeness, comment quality. The systemic quality bars that apply across all agents.
SKILL.md.claude/skills/fleet-quality-enforcement/SKILL.md
GitHubQuality Enforcement — Fleet-Ops Systemic Standards
Review checks the WORK. Quality enforcement checks the CRAFT — whether the work was produced following fleet standards. Commit messages, PR descriptions, artifact completeness, comment structure. These aren't preferences — they're requirements that keep the fleet's output consistent and traceable.
The 5 Quality Standards
1. Conventional Commit Format
Standard: type(scope): description [task:XXXXXXXX]
Enforcement: PreToolUse hook on fleet_commit (exit code 2 = blocked)
What to check during review:
- Every commit follows the format (the hook should catch violations, but verify)
- Types are used correctly:
featfor new,fixfor bugs,refactorfor restructuring - Scopes match the actual changed area (not
feat(update): things) - Task reference matches the actual task
- Each commit is one logical change (no "feat + fix + refactor" in one)
Common violations to flag:
- "feat" for everything (even bug fixes and test additions)
- Missing or wrong scope
- Giant commits mixing multiple concerns
- Task reference doesn't match the PR's linked task
2. PR Structure
Standard: PR body follows the fleet template (generated by fleet_task_complete):
- Summary section with what and why
- Changelog with conventional commit list
- Diff table showing files changed
- Labor stamp with model, effort, backend
- Task reference linking to OCMC
What to check:
- Does the PR tell a reviewer what to look at and why?
- Is the changelog accurate (matches actual commits)?
- Does the labor stamp exist (provenance tracking)?
- Are cross-references present (task URL, Plane issue if applicable)?
Red flag: PR with title only, no body. This means fleet_task_complete's template didn't fire or was bypassed.
3. Artifact Completeness
Standard: Tasks at each stage should have appropriate artifacts:
- Analysis stage: analysis_document with scope, findings, implications
- Investigation stage: investigation_document with options, tradeoffs
- Reasoning stage: plan with requirement reference, target files, steps
What to check:
- Artifact exists for the task's stage history
- Artifact completeness percentage was >80% when stage advanced
- Artifact references specific files and line numbers (not vague descriptions)
Red flag: Task went from analysis to work with no analysis_document. This means stage skipping occurred — flag to accountability.
4. Comment Quality
Standard: Task comments are structured, specific, and actionable:
- Progress comments include: what was done, what's next, blockers
- Contribution comments are typed:
**Contribution (type)** from agent: - Rejection comments specify: what failed, what to fix, evidence
What to check during review:
- Are progress comments present? (hook warns on completion without progress)
- Are contribution comments properly typed? (synergy system depends on this)
- Are rejection comments actionable? (your own hook blocks short rejections)
Red flag: Task with zero comments between accept and complete. No trail of thinking.
5. Trail Completeness
Standard: Every state-modifying tool call should produce a trail event. The PostToolUse hook records to .fleet-trail.log. Board memory records with [trail] tags.
What to check:
- Trail events exist for the task (use
acct_trail_reconstructionif needed) - Key events present: plan_accepted, commits, progress, completion
- Timestamps are sequential (no time travel)
Red flag: Task completed with no trail events. Either the hook isn't firing or the agent bypassed the tool chain.
Enforcement During Review
When you run ops_real_review(task_id), systematically check all 5 standards. Include findings in your approval comment:
## Review: {task_title}
### Quality Standards
✓ Commits: 4 conventional commits, correct types and scopes
✓ PR: full template with changelog, diff table, labor stamp
✓ Artifacts: analysis_document (90%), plan (100%)
⚠ Comments: only 1 progress comment (sparse but present)
✓ Trail: 8 events, chronological, all stages represented
### Work Quality
[your code/design review findings here]
### Decision: APPROVED
Quality standards met. Work satisfies requirement.
Connection to Accountability
Your quality enforcement findings feed accountability's pattern detection:
- If you consistently flag "no progress comments" → accountability detects the pattern
- If you consistently flag "wrong commit types" → accountability tracks the trend
- Your individual reviews become data points for systemic analysis
You enforce per-task. Accountability aggregates across tasks. Together, you maintain fleet quality.