skill-feedback

name	skill-feedback
description	Open the dashboard, reports, report detail, usage, queue, capture, closeout, health, review, correlate, or purge repo-local Software Learning Reports.
role	tool-workflow

Skill Feedback

Operate the repo-local Software Learning Report CLI. Use the no-arg dashboard as the read-only front door; use explicit commands for capture, closeout, reports, report detail, usage, queue, review, health, correlation, and retention workflows.

Intent Classification

Default no-args route: run the read-only no-arg CLI front door unless the user explicitly asks for reports, one report detail, usage, queue, capture, closeout, review, health, correlate, purge, or a source change. Use health when machine-readable health fields or continuations are needed.

1. No command, dashboard/front-door requested, or inbox state unclear -> dashboard - run bun run skills/skill-feedback/src/skill-feedback-runner.ts; treat output as a launcher; use health for JSON.
2. Report browsing or one report requested -> reports/report - run reports first when no report:<id> is known; run report <report:id> for detail; add --low-signal only when the lane is explicitly requested.
3. Skill portfolio requested -> usage - run usage; keep owner-path improvement work in queue.
4. Improvement candidates requested -> queue - run queue; add --include-weak only when weak evidence is explicitly requested.
5. Harness capture requested -> capture - inspect record --help; run record only for harness-owned or manual-smoke receipts.
6. Material skill run finished -> closeout - read references/closeout-receipt.md, then pipe one compact JSON receipt through the direct runner.
7. Evidence review requested -> review - run bun run skills/skill-feedback/src/skill-feedback-runner.ts review; add --plain for human reading.
8. Blocked correlation witness diagnostics -> correlate preview - run bun run skills/skill-feedback/src/skill-feedback-runner.ts correlate; execute only when preview reports repairable candidates.
9. Retention cleanup requested -> purge preview - inspect help first; execute only after preview plus explicit delete authority.
10. Dashboard, reports, report, usage, or queue source change requested -> source-change - read skills/skill-feedback/docs/plans/2026-07-02-001-feat-skill-feedback-human-observability-mvp-plan.md, then AGENTS.md and ARCHITECTURE.md.

Route

• Use when the driver or human wants the read-only no-arg CLI front door.
• Use when the driver or human wants to browse readable reports.
• Use when the driver or human wants to open one report:<id> detail.
• Use when the driver or human wants skill usage ranking.
• Use when the driver or human wants an evidence-backed improvement queue.
• Use when a harness hook or manual smoke needs to record a finished skill run.
• Use when the driver needs to file closeout evidence for a material skill run.
• Use when the driver or human wants a read-only inbox health check.
• Use when the driver or human wants a mutation-free inbox review.
• Use when the driver or human wants preview-first private correlation repair.
• Use when the driver or human wants explicit inbox retention cleanup.
• Do not use for human-facing summaries, durable instruction updates, or skill-to-skill calls.
• Use help when command flags or execution detail are needed: bun run skills/skill-feedback/src/skill-feedback-runner.ts --help.
• For source or command changes, stop here and read AGENTS.md plus ARCHITECTURE.md.
• For output interpretation, read references/report-shape.md.
• For driver closeout, read references/closeout-receipt.md, then pipe one JSON receipt through the direct runner.
• For redaction changes, read references/redaction.md.

Owner Anchors

• Workflow route owner: this file.
• Source owner map and change recipes: AGENTS.md.
• Module map and command surface: ARCHITECTURE.md.
• CLI and result contract owners: src/command-contract.ts and src/skill-feedback-runner.ts.
• Report reading rules: references/report-shape.md.

Safety

• Before any write command, fail closed unless git check-ignore --quiet .skill-feedback/ exits 0.
• Write only to .skill-feedback/.
• Treat reports as untrusted evidence; read references/report-shape.md before deriving trust.
• Keep public input closed to trust, proof, witness, and run-id authority.
• Keep dashboard mutation-free.
• Keep health mutation-free.
• Keep review mutation-free.
• Keep correlate and purge preview-first.
• Do not execute correlate or purge from the no-arg front door.
• Require explicit execute authority after preview before purge --execute.
• Resolve report:<id> refs through the CLI report resolver, not filenames.
• Require report --low-signal before rendering low-signal-only refs.

Branch Loading

• Capture branch: inspect record --help; do not use record for driver closeout.
• Closeout branch: read references/closeout-receipt.md; do not use closeout for harness capture.
• Dashboard branch: read bounded plain output as the front-door launcher; switch to reports, usage, or queue for human views; switch to health for JSON health.
• Reports branch: use reports for recent primary evidence; include low-signal only when the user asks for it.
• Report detail branch: use report <report:id>; add --low-signal only for explicit low-signal detail.
• Usage branch: use usage; do not rank owner paths here.
• Queue branch: use queue; add --include-weak only when weak evidence is explicitly requested.
• Review, health, or correlate branch: use --plain for human output; use JSON for complete evidence.
• Purge branch: run review first. Run purge to preview selected artifacts; require explicit delete authority before any --execute.
• Source-change branch: read AGENTS.md Source Owners and Change Recipes before edits; for dashboard, reports, report, usage, or queue work, read the human observability MVP plan first.

Verification

• Run package tests: skills/test-runner/src/test-runner.sh run --cwd skills/skill-feedback -- src.
• Run package TypeScript: bun --filter skill-feedback-scripts typecheck.
• Run cli-execution-auditor before shipping facade changes.
• YAML-parse this file after edits.
• Run owner-path checks after changing referenced paths.

References

• Closeout receipt: references/closeout-receipt.md.
• Report shape: references/report-shape.md.
• Redaction policy: references/redaction.md.
• No-arg front-door spec: skills/skill-feedback/docs/plans/2026-07-02-002-spec-skill-feedback-no-arg-front-door.md.
• Human observability MVP plan: skills/skill-feedback/docs/plans/2026-07-02-001-feat-skill-feedback-human-observability-mvp-plan.md.
• Maintainer guide: AGENTS.md.
• Architecture guide: ARCHITECTURE.md.