git:20260707.48f8797 to git:20260707.8f5c5dd

5 added, 4 removed. Audit A to A.

# CLAUDE.md

> This file stacks on top of the workspace root at `C:\Code\GitHub\`:

> - Root [`CLAUDE.md`](../../CLAUDE.md) -- voice, rules, routing map, references, skills, slash commands, conventions.

> - Root [`MEMORY.md`](../../MEMORY.md) -- live facts across repos.

> - Root [`STATUS.md`](../../STATUS.md) -- live PR/CI/security dashboard.

> - [`.claude/resources/`](../../.claude/resources/README.md) -- deep reference for collaboration, workflow, git, OSS, debugging, voice.

> Read those first. The guidance below only adds **repo-specific context** -- it does not override anything in the root.

## Project

Conformance suite for Agent Skills (SKILL.md): lints skills against the agentskills.io spec today; roadmap is headless cross-runtime execution tests + a public caniuse-style compatibility matrix. Public OSS, npm package `@sagargupta1610/skillcheck`.

## Stack

- - **Language**: TypeScript 5 (strict, ESM, NodeNext)

+ - **Language**: TypeScript 6 (strict, ESM, NodeNext)

- **Framework**: none -- CLI via commander, build via tsup

- **Database**: none

- **Package manager**: pnpm

- **Deploy target**: npm registry (CLI); matrix site later (Vercel)

## Run

```

pnpm install

pnpm dev # tsup --watch

pnpm build # dist/

node dist/cli.js lint fixtures/valid-skill

```

## Test

```

pnpm test # vitest

pnpm lint # biome

pnpm typecheck

```

## Entry points

- `src/cli.ts` -- commander CLI (`skillcheck lint`)

- `src/index.ts` -- library exports (lintSkillDir, parseSkillMd, types)

## Key files

- - `src/lint/rules.ts` -- the "brain": all lint rules, spec fields, known client extensions, limits

+ - `src/lint/rules.ts` -- the "brain": all lint rule implementations and limits

+ - `src/extensions.ts` -- SPEC_FIELDS + KNOWN_EXTENSIONS typed registry (spec fields, client extension fields)

- `src/lint/parse.ts` -- SKILL.md frontmatter/body splitter (never throws)

- `fixtures/` -- valid-skill and broken-skill lint fixtures used by tests and CI smoke test

## Gotchas

- The Agent Skills spec is UNVERSIONED -- when rules change upstream, pin claims to a spec-repo git SHA in the finding message/docs.

- - Severity model is two-tier by design: error = skills-ref strict parity, warning = client-guide lenient parity. `--strict` escalates warnings for `name-dir-match`, `extension-field`, `unknown-field`. Don't collapse the tiers.

- - `KNOWN_EXTENSIONS` in `src/lint/rules.ts` tracks Claude Code extension fields; update it when Claude Code ships new frontmatter keys.

+ - Severity model is three-tier by design: error = skills-ref strict parity, warning = beyond-parity checks + client-guide downgrades, info = advisory (never gates CI). `--profile strict` is the default; `--profile lenient` downgrades `name-too-long`, `name-dir-mismatch`, `unknown-frontmatter-field` to warnings (`--strict` is a deprecated alias for the default). Don't collapse the tiers.

+ - `KNOWN_EXTENSIONS` in `src/extensions.ts` tracks client extension fields (Claude Code, Cursor, Copilot, OpenCode); update it when runtimes ship new frontmatter keys.

- Name collision landscape: getskillcheck.com is an unrelated paid static-analysis product; npm unscoped `skillcheck`/`skill-check` are taken. This package is scoped `@sagargupta1610/skillcheck` on purpose.

## Repo-specific rules

- v0.2+ runtime execution work must NEVER run agent CLIs outside Docker containers, and user API keys are supplied by env var only -- never stored.