claude-standardslisted

# Claude Standards Audit Audits the five classes of Claude Code assets in this repo against the best-practice reference documents in this directory. Default mode is **report only**. With `--fix`, applies a narrow set of conformance-only changes — never feature additions or removals. ## Asset inventory | Asset class | Location | Standards reference | |-------------|----------|---------------------| | Skills | `.claude/skills/*/SKILL.md` (plus `.claude/skills-global/*/SKILL.md` if the repo has one) | [STANDARDS_SKILLS.md](STANDARDS_SKILLS.md) | | Slash commands | `.claude/commands/*.md` | [STANDARDS_COMMANDS.md](STANDARDS_COMMANDS.md) | | Subagents | `.claude/agents/*.md` | [STANDARDS_SUBAGENTS.md](STANDARDS_SUBAGENTS.md) | | Hooks | `.claude/settings.json` + `.claude/hooks/**` | [STANDARDS_HOOKS.md](STANDARDS_HOOKS.md) (canonical source: the `audit-hooks` skill's `BEST_PRACTICES.md`, installed alongside this skill) | | MCP servers | `config/mcp_library.json` + `mcp_servers/` (if present) | [STANDARDS_MCP.md](STANDARDS_MCP.md) | Each `STANDARDS_*.md` file holds the rules. Keep those files as the single source of truth; do not inline rules here. ## References Cross-cutting reference docs live under `references/` and are loaded only when the task calls for them (progressive disclosure): - [`references/EVALS.md`](references/EVALS.md) — how to evaluate a prompt systematically. Read when iterating on a prompt that ships inside a skill, subagent, hook, or agent loop. - [`refer