- name
SKILL.md
Файл навыка остаётся в репозитории издателя.
Вставьте скопированную команду в терминал. Она установит этот SKILL.md в выбранного агента.
npx skills add BlackBeltTechnology/pi-agent-dashboard --skill code-quality -a cursor -yВставьте скопированную команду в терминал. Она установит этот SKILL.md в выбранного агента.
npx skills add BlackBeltTechnology/pi-agent-dashboard --skill code-quality -a claude-code -yВставьте скопированную команду в терминал. Она установит этот SKILL.md в выбранного агента.
npx skills add BlackBeltTechnology/pi-agent-dashboard --skill code-quality -a codex -yВставьте скопированную команду в терминал. Она установит этот SKILL.md в выбранного агента.
npx skills add BlackBeltTechnology/pi-agent-dashboard --skill code-quality -a grok -yОтправьте скопированное сообщение в чат Grok Bot. Бот загрузит этот SKILL.md с GitHub и добавит его к навыкам, общим для всех ваших ботов.
Install this skill: https://github.com/BlackBeltTechnology/pi-agent-dashboard/tree/HEAD/.pi/skills/code-qualityОт издателя
Analyze → fix → test, drivable by the goal-plugin's judge loop. The skill is the HOW; the goal is the WHEN-TO-STOP (judge reads one exit code).
flowchart TD
Q{Driving a goal /<br/>cleaning your own work?}
Q -->|default| C[CHANGED-FILES MODE<br/>goal loop · git diff vs develop<br/>surgical, low blast radius]
Q -->|explicit cleanup| W[WHOLE-REPO MODE<br/>human-gated · one package,<br/>one rule-group per turn]
C --> CC[npm run quality:changed]
W --> WW["biome lint <pkg> --only=<group>"]
When unsure, use changed-files mode. Never turn whole-repo autofix loose inside a goal loop.
One command, one exit code:
npm run quality:changed
# = biome check --changed --error-on-warnings --write (safe-fix the diff)
# && node scripts/tsc7.mjs --noEmit (type gate, TypeScript 7)
# && npm test (test gate)
--changed compares the committed branch state against develop (the repo's
integration branch, set as vcs.defaultBranch in biome.json). Uncommitted or
untracked files are NOT scoped by --changed — commit your work first, or pass an
explicit path.
1. ANALYZE
npm run quality:changed # or: biome lint . --reporter=json (whole-repo)
Read the JSON. Group issues: safe-fixable vs unsafe vs manual.
2. FIX (surgical — only files in the diff)
- Safe fixes are already applied by `--write`. Confirm they're sane.
- Type errors + unsafe-rule + manual issues: edit by hand, minimally.
- DO NOT run `--unsafe` in the loop. Surface unsafe fixes as a report.
- DO NOT touch files outside the diff (AGENTS.md surgical rule).
3. TEST (the gate)
node scripts/tsc7.mjs --noEmit && npm test
- Red? REVERT this fix batch. Never stack broken autofixes.
- Green + 0 issues on scope? Done. Else loop.
biome check --write applies safe fixes only. Some of this repo's enabled
rules are unsafe and need a human (or --unsafe, never in a loop):
--write): useConst, useImportType, noUnusedImports.useTemplate, useOptionalChain,
noUnusedVariables (and any rule Biome marks FIXABLE but doesn't apply under
plain --write).If --write reports a warning as FIXABLE yet leaves it, the fix is unsafe.
Apply it manually with judgment; do not blanket --unsafe.
--changed). Whole-repo only when
explicitly asked, and then scoped to one package + one rule-group per turn.node scripts/tsc7.mjs --noEmit (the TypeScript 7 type-check) + npm test after every fix batch; revert on red.Biome lints whole files, not diff lines. Touch one line in a legacy file → all
its warn-tier issues surface. Default policy: grandfather — fix only the
diff, // biome-ignore unavoidable legacy lines, leave the rest. Do NOT rewrite
a whole legacy file to chase warnings (violates surgical rule) unless the user
asks for boy-scout cleanup.
Daily driver (changed files):
/goal "Use the code-quality skill in changed-files mode. Done when
`npm run quality:changed` exits 0. Pause and report if a fix needs a
non-mechanical judgment call. Never edit files outside the diff."
Scoped cleanup (one package):
/goal "Use the code-quality skill, whole-repo mode, scoped to <pkg>. Drive
`biome lint <pkg> --only=<group>` to 0 while keeping tests green. One
rule-group per turn; stop after each so I can review the diff."
Severity = gate strength. warn shows but exits 0 (soft); error exits
non-zero (hard). CI runs plain biome lint . → only error-tier blocks. The
goal loop adds --error-on-warnings → warn+error both block on YOUR files. So
legacy warn-tier noise never fails CI, but new/touched code is held to the full
bar. See docs/code-quality.md for the tier ladder + graduation rules.
npm run quality:changed # the oracle: fix diff + tsc + test (goal loop)
npm run fix:changed # safe-fix the diff only, no test gate
npm run lint:biome # analyze whole repo (no fixes)
npm run quality:report # whole-repo GitHub-annotation report (advisory)
biome lint <path> --only=<group>/<rule> --reporter=json # targeted cleanup
Drive static-analysis code quality in pi-agent-dashboard with Biome (analyze → fix → test), in changed-files or whole-repo mode. Use when asked to "improve code quality", "lint and fix", "clean up war
---name: code-qualitydescription: Drive static-analysis code quality in pi-agent-dashboard with Biome (analyze → fix → test), in changed-files…---# .pi/skills/code-quality/SKILL.mdОбзор
Drive static-analysis code quality in pi-agent-dashboard with Biome (analyze → fix → test), in changed-files or whole-repo mode. Use when asked to "improve code quality", "lint and fix", "clean up warnings", "fix Biome issues", "run static analysis", or when setting a code-quality goal. Skip for one-line edits.
FAQ
Скачайте .pi/skills/code-quality/SKILL.md у издателя или скопируйте путь и откройте репозиторий. PlugMCP только показывает навык в каталоге и не хранит файл.
Издатель хранит его в https://github.com/BlackBeltTechnology/pi-agent-dashboard.
Claude, Cursor, Codex и Grok Build могут читать SKILL.md, который вы добавляете из репозитория издателя. Grok Bot устанавливает его, когда вы отправляете в чат ссылку на папку навыка на GitHub.