메뉴
Format code, run linters and static analysis, and find unused code (ESLint + Knip). Use when the user asks to fix lint, format code, run formatters, find dead code, unused exports, run knip, fix style, tidy code, or run eslint --fix.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Upgrade outdated npm dependencies in the pnpm monorepo (recursive update, latest, targeted packages). Use when the user asks to upgrade dependencies, bump packages, refresh lockfile, pnpm update, outdated packages, latest versions, or dependency refresh—not monorepo release versions.
Manage Architecture Decision Records (ADRs). Use this to initialize, create, list, and link ADRs to document architectural evolution. Requires 'adr-tools' to be installed.
Use at the end of a coding session to summarize outcomes, failures, inefficiencies, and root causes, then output a concise postmortem report with ranked improvements for next time. Output only in chat; do not edit project files unless the user explicitly asks. Skip nit-picks and one-off mistakes.
Interactively bump package versions for the parser package. Use when the user asks to bump version, increment version, set release version, or release dbt-artifacts-parser.
Regenerates TypeScript types from dbt artifact JSON schemas. Use when the user asks to update parsers, sync with upstream, refresh types, or regenerate parser types from existing schemas.
Interactive deep research and decision support: frame the real problem (XY-aware), ask exactly 10 multiple-choice questions one at a time, then produce a rigorous comparative evaluation (default 5 approaches, 0–100 scores) and recommendation. Use when the user wants structured discovery before committing to a solution, a scored comparison of approaches, or to avoid jumping straight to an answer—especially for architecture, strategy, or high-stakes trade-offs.
| name | lint-fix |
| description | Format code, run linters and static analysis, and find unused code (ESLint + Knip). Use when the user asks to fix lint, format code, run formatters, find dead code, unused exports, run knip, fix style, tidy code, or run eslint --fix. |
| compatibility | Requires pnpm from the repository root; expects `package.json` scripts and repo configs (`eslint.config.mjs`, `knip.json`). The **default** path assumes **`pnpm install`** has been run so **`@trunkio/launcher`** is present; use **`pnpm format:without-trunk`** / **`pnpm lint:without-trunk`** only as an escape hatch for **broken or minimal** environments where `trunk` cannot run. |
Activate this skill when the user says or implies:
| Approach | Feasibility | Maintainability | Fit for agents | Score |
|---|---|---|---|---|
| ESLint only for “unused” | High | High | Misses cross-package unused exports | 55 |
| Knip only | High | Medium | Misses many intra-file issues ESLint covers | 60 |
| Format + ESLint + Trunk + Knip (layered) | High | High | Clear order, matches repo scripts | 92 |
| Bundle / tree-shake reports as primary | Medium | Low | Indirect, poor for “delete this symbol” | 40 |
| IDE-only cleanup | Low | Low | Not reproducible in CI or agents | 25 |
Use the layered flow: format for consistent whitespace, ESLint for type-aware rules and unused bindings, Trunk for shared checks (when available), Knip for workspace graph (unused exports, files, dependencies).
Default (this monorepo): From the repository root, run pnpm install so @trunkio/launcher is installed. Then use pnpm format / pnpm lint, or pnpm exec trunk check, pnpm exec trunk fmt, for full parity with CI. pnpm puts node_modules/.bin on PATH when running package scripts, but a bare interactive shell often does not—so command -v trunk / which trunk are not authoritative; do not use them alone to decide if Trunk is usable.
Optional checks (from repo root): test -x node_modules/.bin/trunk, or pnpm exec trunk version.
| Situation | Format | Lint (full, no fix loops) |
|---|---|---|
| Use pnpm scripts (launcher present) | pnpm format | pnpm lint |
Launcher missing / cannot run trunk | pnpm format:without-trunk (same as pnpm format:eslint && pnpm format:prettier) | pnpm lint:without-trunk (same as pnpm lint:eslint && pnpm lint:stylelint && pnpm knip) |
Do not run pnpm format or pnpm lint when the launcher cannot execute trunk (e.g. missing node_modules or failed install): both scripts invoke trunk first and will fail immediately. The *:without-trunk scripts are an escape hatch, not the normal workflow here.
Coverage gap (no Trunk): .trunk/trunk.yaml enables additional checks (e.g. markdownlint, taplo, shellcheck, actionlint, Trivy, OSV, yamllint) that the no-Trunk path does not run. The no-Trunk path matches most TypeScript / React / CSS / dead-code work and aligns with agent gates in AGENTS.md (lint:report, knip) but is not CI-identical. Before merge or when editing markdown, YAML, shell, or workflow files, run pnpm lint in an environment with Trunk when possible.
CSS note: this parser repository does not run Stylelint as part of its normal gates; pnpm lint:report is ESLint-only and writes lint-report.json.
sonarjs/no-duplicate-string (threshold is typically 3 repeats). Fix by hoisting const descriptions next to existing option constants. Large functions may hit sonarjs/cognitive-complexity or sonarjs/cyclomatic-complexity — extract small pure helpers rather than disabling rules.eslint-plugin-import now enforces import/no-cycle, import/no-unresolved, and import/no-useless-path-segments on TypeScript files. Resolve cycles by moving shared types/helpers to a lower layer or introducing a facade, not by disabling the rule.@eslint-community/eslint-plugin-eslint-comments enforces eslint-comments/no-unused-disable and eslint-comments/disable-enable-pair; remove stale directives before re-running lint.__ must use kebab-case segments only (e.g. __search-row, not __searchRow). If Trunk reports stylelint/selector-class-pattern, rename the class in both TSX and CSS.*.md files must not link to missing paths. Run pnpm exec trunk check path/to/file.md -y on edited docs before merge.pnpm format can fail on ESLint even after Trunk fmt succeeds — the script runs format:eslint second; read the ESLint output and fix before re-running format.pnpm format / pnpm lint)pnpm format (trunk fmt → eslint . --fix → prettier --write .).pnpm lint (lint:trunk → lint:eslint → lint:stylelint → knip), or a slimmer pass: pnpm lint:trunk && pnpm lint:eslint && pnpm lint:stylelint (omit Knip if not needed yet).packages/** TypeScript modules changed, run pnpm lint:eslint and clear import-cycle/path issues before spending time in pnpm test:e2e.pnpm lint:eslint checkpoint, if the touched files include parser source, generated types, package exports, or shared TypeScript configuration, run pnpm build from the repo root. If it fails, switch to .claude/skills/build-fix/SKILL.md and follow its fixer loop before claiming the lint session is complete.*:without-trunk)Format first — pnpm format:without-trunk (or pnpm format:eslint && pnpm format:prettier). Do not use pnpm format.
Lint — pnpm lint:without-trunk (or the same chain manually). Do not use pnpm lint.
Dead code — pnpm knip is already at the end of lint:without-trunk; treat pnpm knip:fix as review-only (it can remove exports and files).
Do not treat knip --fix as a formatter. Prefer manual or reviewed edits for removals.
eslint.config.mjs — TypeScript recommended + @typescript-eslint/no-unused-vars, @typescript-eslint/no-unused-private-class-members on production packages/**/*.ts(x) (excludes *.test.* and e2e specs), plus eslint-plugin-import (cycles/unresolved/useless segments), eslint-plugin-playwright on e2e specs, and eslint-comments guardrails for disable directives. React / a11y / hooks for web TSX. Fix via pnpm format:eslint and pnpm lint:eslint.knip.json — per-workspace entries (Vitest, parser, core, CLI, web + Vite/Playwright), ignoreExportsUsedInFile, targeted ignoreIssues / ignoreFiles / ignoreDependencies / ignoreBinaries. Decision record: docs/adr/0005-knip-and-eslint-layers-for-monorepo-dead-code-detection.md.Run from the repository root.
| Step | Command | Purpose |
|---|---|---|
| Format (launcher path) | pnpm format | trunk fmt → eslint . --fix → prettier --write . |
| Format (escape hatch) | pnpm format:without-trunk | eslint . --fix → prettier --write . only |
| Lint full (launcher path) | pnpm lint | lint:trunk → lint:eslint → lint:stylelint → knip |
| Lint full (escape hatch) | pnpm lint:without-trunk | lint:eslint → lint:stylelint → knip (no Trunk) |
| ESLint report (agents) | pnpm lint:report | Writes lint-report.json; ESLint-only; must exit 0 for green |
| Dead code (analyze) | pnpm knip | Unused exports/files/deps; must exit 0 when clean |
| Dead code (assisted edit) | pnpm knip:fix | Review diff before commit |
Typical fix sequence — launcher present:
pnpm format && pnpm lint:trunk && pnpm lint:eslint && pnpm lint:stylelint && pnpm knip
(Equivalent high-level: pnpm format then pnpm lint.)
Typical fix sequence — escape hatch:
pnpm format:without-trunk && pnpm lint:without-trunk
If violations remain:
knip.json when needed (shell-only scripts, dynamic imports, published API).pnpm format (or at least ESLint + Prettier), then pnpm lint:eslint, pnpm knip, pnpm lint:report.pnpm build after the first green pnpm lint:eslint pass; if build breaks, use the build-fix skill loop before final verification.pnpm format:without-trunk, pnpm lint:without-trunk, pnpm lint:report.pnpm lint:stylelint in verify if you did not run full pnpm lint / pnpm lint:without-trunk..claude/, .github/workflows/, .trunk/, or other files Trunk owns, include pnpm lint:trunk (or full pnpm lint) when Trunk is available—pnpm lint:report does not run markdownlint.pnpm lint from a repo with pnpm install completed when possible.Policy: Fix root causes first; avoid blanket eslint-disable or widened ignore rules. If suppression is unavoidable, keep it narrowly scoped and do not require inline justification comments.
The verifier subagent owns step ordering, parallelism, the step 7 stability loop (stability_iterations, cap 3, and which gates rerun after normalization)—see .claude/agents/verifier.md Step 7 — normalization stability loop (cap: 3).
Use the Order rule above in this skill:
pnpm format then pnpm lint (equivalent high-level sequence; see Commands for this repo and Typical fix sequence — launcher present for the decomposed chain).pnpm format:without-trunk then pnpm lint:without-trunk when Trunk/launcher cannot run—see Path B under Order rule; do not use pnpm format / pnpm lint on Path B.When pnpm format / pnpm lint (Path A) or the Path B equivalents still leave non-empty git status --porcelain, the verifier only re-runs steps 1–3 (lint report gate, unit test gate, coverage report gate)—not build, pack smoke, or CodeQL. See the verifier’s Normalization: what reruns automatically vs what does not.
Confirm at minimum:
pnpm lint:report exits 0 (ESLint-only; see Commands table — it does not run markdownlint or other Trunk linters).pnpm knip exits 0A full verifier run completes step 7, which includes full lint (Path A: pnpm lint includes pnpm lint:trunk and Knip). If you only run the minimum above and the change set touches Markdown, .claude/, workflows, or .trunk/, also run pnpm lint:trunk (or full pnpm lint) before claiming merge-ready so Trunk-owned checks match pre-push parity.
When using the escape hatch (*:without-trunk), also run pnpm lint:eslint and pnpm lint:stylelint (or a single pnpm lint:without-trunk) so non-ESLint issues (especially CSS) are not missed. Trunk-only checks (markdownlint, actionlint, etc.) still require pnpm lint:trunk / pnpm lint when the launcher is available.
Re-run the relevant commands after fixes before marking the step complete.
If scripts differ, prefer the repo’s package.json format / lint / knip (or equivalent) names. Keep the same order: format → lint fixes → dead-code graph tool if present. If Trunk-equivalent tooling is optional there too, document a no-Trunk subset the same way.