AI Agent & Automation Workflows

Use when auditing a CLAUDE.md or AGENTS.md file. Scores six metrics (0-100), cites issues, suggests fixes. Codebase scan first, falls back to file-only when no codebase available.

Use when iteratively raising a CLAUDE.md or AGENTS.md to a target audit score. Spawns blind reviewer + fixer subagents per loop (target hidden, prevents score gaming).

Use when mapping where CLAUDE.md or AGENTS.md files should exist. Scans for non-obvious complexity, evaluates existing files, identifies gaps. Run before agents-md-audit on each flagged file.

Use when writing a client-facing email summarizing project work. Reads git history + code, drafts a tiered outline by client impact, outputs markdown with optional HTML/screenshots.

Use when finding architectural friction in a codebase. Surfaces shallow modules, proposes refactors for testability/AI-navigability, grills the design. Uses CONTEXT.md + ADRs.

Use when squashing commits on the current branch into fewer logical commits. Triggers: squash, clean up commits, combine commits, tidy history before push or PR.

Use when evaluating how complex a task is before deciding on an approach. Scores scope, clarity, risk, and novelty to produce a complexity tier (trivial, moderate, complex). Use this whenever the user asks how hard something is, wants a complexity check, or before starting unfamiliar work.

Use when stress-testing a plan against the existing domain model and documented decisions. Challenges terminology, cross-references code, and updates CONTEXT.md and ADRs inline as decisions crystallize.

Use when breaking a plan, spec, or PRD into independently-grabbable GitHub issues using vertical-slice tracer bullets. Each issue is a thin end-to-end slice that can be implemented and verified on its own.

Use when breaking a plan, spec, or PRD into local markdown task files as vertical-slice tracer bullets. No GitHub. Each slice gets its own file with status/blocked-by/type frontmatter.

Use when turning a conversation or idea into a PRD file. Synthesizes existing context into a structured product requirements document with user stories, module design, and testing decisions. Writes to plans/prd.md for use with the ralph-loop skill. Do NOT interview the user — just synthesize what you already know.

Use when beginning work on a NEW task that hasn't been explored yet. Assesses complexity, then routes to the right execution depth — trivial tasks get solved immediately, moderate ones are broken into steps, complex ones get a plan first. Do NOT use when the task has already been discussed, explored, or planned in the current conversation — in that case just execute directly.

Use when locking down user paths as E2E regression tests. Explores the actual UI in a browser, proposes user flows, writes automated E2E tests (Playwright/Cypress/etc.), and runs them until green. Use this whenever the user says test flow, test path, test user journey, cover this flow, or wants regression tests for user-facing behavior.