Qv Quality Reporting

Quality Reporting

Turn the repository's deterministic quality evidence into a short, actionable review. This skill owns contextual grouping and triage; it does not replace the underlying findings or mutate Asana without explicit approval.

Default Workflow

1. Run pnpm quality:reporting from the repository root.

2. Stop if the command fails, any detector reports diagnostics, or resolutions are withheld. Explain the incomplete evidence instead of proposing progress.

3. Read .quality/triage.json, references/triage-config.json, and only the source or architecture files needed to understand the leading candidates.

4. Form cohesive remediation groups. Start with deterministic candidates, then:

   • combine candidates only when they share one root cause and fit one focused PR;
   • split a candidate when it contains independent remediations;
   • preserve every covered finding fingerprint;
   • apply approved merge/split overrides from the config before new judgment.

5. Assign ownership and P1-P4 priority using the config plus evidence in the triage report. Never derive priority from finding count alone.

6. Build the review batch using references/ticket-template.md:

   • On the initial backlog review, begin with the deterministically ranked candidates.
   • On recurring runs, examine all new, worsened changed, and resolved lifecycle evidence before unchanged existing debt, regardless of severity. Skip improved-only and unchanged groups unless the user asks for backlog review.
   • Generate provisional proposals in pages, reconcile each page read-only, discard exact-existing and unchanged outcomes, and continue until ten actionable proposals remain or candidates are exhausted. Ambiguous matches count as actionable decisions. Never cap the source candidates before this backfill step. Write the final machine batch to .quality/proposals.json. Retain unreviewed candidates for later pages; do not discard them.

7. Reconcile each provisional page read-only with:

   node .agents/skills/qv-quality-reporting/scripts/asana-quality.mjs \
     --proposals .quality/proposals.json \
     --report .quality/triage.json

   Suppress exact duplicate creations and unchanged actions, surface ambiguous matches, and show the resulting final batch in chat as a numbered, concise review. If local Asana access is unavailable, label the proposals unreconciled and do not apply them until reconciliation succeeds.

8. End the initial run by asking the user which proposal IDs to approve. Do not use --apply, even if the user previously authorized analysis.

9. After the user explicitly approves IDs, rerun the helper for only that subset without --apply and show the exact actions. Stop if the match is ambiguous or the result differs materially from what the user approved. Otherwise run:

   node .agents/skills/qv-quality-reporting/scripts/asana-quality.mjs \
     --proposals .quality/proposals.json \
     --report .quality/triage.json \
     --approve <comma-separated-proposal-ids> \
     --apply

   Never infer approval for proposals the user did not name.

Review Shape

For each proposal, show:

• proposal ID, P1-P4 priority, primary team, and collaborators;
• a two-to-four sentence explanation of what is wrong and where;
• one focused remediation boundary;
• compact evidence: paths, rules, severity, threshold excess, and Git activity;
• lifecycle action: create, comment on a materially worsened group, or review an apparent resolution.

Prefer source links in chat when a concrete line is available. Keep the deterministic finding list in .quality/report.md as the evidence trail.