| name | linting-llm-configs |
| description | Validates AI guidance files with claudelint and agnix. Use when checking SKILL.md, CLAUDE.md, prompts, hooks, or MCP config for structural issues. |
Linting LLM Configs
Validation-only companion to ai-tuning. Keep remediation focused on lint findings; use ai-tuning when the task is about condensing, deduplication, or instruction design.
Usage
Use this skill after editing AI config files or when CI reports config-lint failures.
When to use
| Situation | Validator |
|---|
Claude Code structure, hooks, MCP, CLAUDE.md | claudelint |
SKILL.md, AGENTS.md, cross-tool config checks | agnix |
| Mixed repo or release gate | Run both |
Workflow
- Identify the changed files and choose
claudelint, agnix, or both.
- Run the narrowest command that covers the edited surface.
- Fix structure, naming, or duplication issues in the source files.
- Re-run the same validators until they pass.
- If the content is verbose but lint-clean, switch to
ai-tuning for condensation work.
Commands
| Scope | Command |
|---|
| Full Claude Code lint | claudelint check-all |
| Strict repo validation (matches CI) | uv tool run "claudelint@0.3.3" --strict . |
| Validate skill files | claudelint validate-skills --path . |
| Validate hooks | claudelint validate-hooks |
| Validate MCP config | claudelint validate-mcp |
| Cross-tool lint | agnix . |
| Safe autofix pass | agnix --fix-safe . |
| Claude Code target | agnix --target claude-code . |
| Cursor target | agnix --target cursor . |
Guardrails
- Keep
SKILL.md files short, trigger-driven, and repo-relevant.
- Remove duplicated rules instead of copying the same guidance into multiple skills.
- Preserve valid frontmatter and kebab-case skill names.
- Do not invent local commands; prefer repo or tool-native commands already in use.
Examples
- "Lint this SKILL.md after edits" -> run
agnix . plus targeted claudelint validate-skills --path .
- "Why did config lint CI fail?" -> reproduce with
uv tool run "claudelint@0.3.3" --strict .
- "Make these skill files shorter" -> switch to
ai-tuning after validators are green
References