Skip to main content

Project init commands

In Claude Code these are slash commands. In Codex use the corresponding generated skill; source-command-health-check disambiguates the project setup check from AWS health checks.

/init-project​

Detect the repository's language, framework, commands, and existing layout. Create missing instructions, documentation, skills, and supported host integration files without overwriting handwritten content.

/sync-docs​

Audit maintained documents against source and configuration; update stale commands, counts, paths, and architectural statements. Keep one authoritative statement for details that change frequently.

/add-adr​

Create a numbered decision record with context, alternatives, decision, and consequences. Link superseded records when appropriate.

/add-module​

Add a module directory and scoped instructions, then update architecture documentation to describe its role and dependencies.

/add-runbook​

Create a runbook with prerequisites, operations, expected output, verification, and recovery steps.

/add-reference-doc​

Add implementation reference material for selected layers under docs/reference/ and link it from the owning instructions.

/generate-readme​

Describe the project's actual purpose, installation, usage, development commands, and contribution path. Follow the requested language.

/generate-changelog​

Record release changes under the project's versioning and changelog conventions. Describe shipped behavior and avoid inventing release dates.

/health-check​

Validate files, hooks, permissions, instruction quality, and configuration. Report the rubric, evidence, and fixes; a health score is not a replacement for the required test suite. Detects pre-v2.3 hook/agent contracts (hooks reading legacy env vars instead of stdin JSON, .yml/.yaml agent files Claude Code never loads) and recommends /migrate-hooks.

/migrate-hooks​

Migrate hooks, settings.json, and agents generated before v2.3 to the current Claude Code contract: hook events arrive as JSON on stdin (not legacy $TOOL_INPUT_PATH/$EVENT/$MESSAGE environment variables), only exit 2 blocks a tool call, and subagents must be Markdown files with YAML frontmatter (.yml/.yaml agent files are never loaded). Keeps a backup; supports --dry-run to report without writing. This only affects files already generated into a consuming project — a plugin update alone cannot fix them there.