skills

Produce a thorough, self-contained implementation plan for a coding task. Explores the codebase read-only, considers alternatives, enumerates risks, and writes a plan suite (an index plus one or more chunk files) that can be executed in a fresh Claude Code session via /execute-plan. Detects under-decomposed tasks to offer splits. Use this when the user invokes /deep-plan or asks for a deep plan before implementation.

Execute a chunk file produced by /deep-plan. Reads the specified chunk, validates suite dependencies via the suite index, confirms files and order with the user, implements in sequence, verifies each acceptance criterion, and prompts for the final status update. Use when the user invokes /execute-plan <path>.

Guided onboarding for newcomers to a codebase. Three phases — gate on CLAUDE.md quality, run a conversational tour driven by CLAUDE.md and the docs it points at, then write one contrived hands-on task to `.claude/onboarding-task.md` for the newcomer to implement themselves on a sandbox branch. Read-only on the codebase during the tour; writes exactly one artifact. Aborts with a pointer to `/claude-md-refactor` (bootstrap mode) if CLAUDE.md is missing, and warns + offers an override if it looks thin. Use when the user invokes /onboarding, says "I'm new to this codebase", "help me onboard", "give me a tour of this repo", "show me around", or otherwise asks to be ramped up on an unfamiliar project.

Read-only status view for plan suites in the current project. Detects the active suite (one with chunks not yet done) and shows status, dependencies, and what's runnable next. Use when the user invokes /plan-tracker or asks "what's the state of the plan" / "what's next in the plan" / "what's left".

Read-only audit of a finished (or near-finished) plan suite. Compares the suite's index.md goal and per-chunk acceptance criteria against the actual git diff since the suite was created, and reports goal alignment, coverage gaps, drift (changes not traceable to any chunk), and status-sanity issues (chunks marked done whose criteria don't appear satisfied). Use when the user invokes /review-plan, or asks "is the plan actually done?", "did we deliver what the plan promised?", "is the suite status honest?", or wants a final check before declaring a plan suite complete.

Configure where plan suites live and whether the plans directory is gitignored. Writes `.claude/plans-config.json` with `plansDir` (default `.claude/plans`) and `gitignore` (default `true`) keys. Existing planning skills (`/intake`, `/deep-plan`, `/simple-plan`, `/plan-tracker`, `/execute-plan`, `/review-plan`) read this config to resolve the plans directory, falling back to `.claude/plans` when the config is absent. Use when the user invokes `/setup`, asks to "configure the plans directory", "change where plans live", "set up the plans skill config", or wants to opt in/out of gitignoring the plans directory.

Lightweight planner for straightforward, mechanical tasks where no architectural decision is involved — single-file bugfixes, renames, flag additions, small refactors confined to one or two files. Runs a compressed flow (restate → brief explore → specify → risk & verify → persist) and writes a single `simple-plan.md` file into a new or adopted suite directory under the configured plans directory (default `.claude/plans/`). When invoked on a suite dir that already contains a `context.md` (typically from `/intake`), reads it and writes output into the same dir; otherwise mints a fresh suite dir. Does NOT integrate with `/execute-plan` or `/plan-tracker`; the user implements the plan directly. Use when the user invokes /simple-plan, when `/intake` recommended `/simple-plan` (no ADR-gate criterion fired), or when the user asks for a quick/lightweight plan for a small task.

Front-door planning skill that gathers context before a coding task is planned. Requires the user to state their intent first; if /intake is invoked with no task, it asks for one and stops. Order is: restate → ask the 3 core questions (intent, scope, anchor) → explore the codebase bounded by those answers to resolve anything code can settle → continue asking only the follow-ups code couldn't answer (≤5 questions total across core + escalate, one at a time via AskUserQuestion, recommended-answer option first). Recommends either /simple-plan or /deep-plan via an ADR-gate (hard-to-reverse / surprising-without-context / real-tradeoff). Writes a single `context.md` artifact into a new suite directory under the configured plans directory (default `.claude/plans/`) and prints a handoff command for the user to run in a new session. Use when the user invokes /intake, asks to "gather context before planning", or wants a checkpoint before /deep-plan or /simple-plan.

Refactor, audit, or bootstrap a repository's CLAUDE.md so it stays lean and acts as an index of pointers to /docs/ files instead of bloating session context. Use this skill whenever the user mentions their CLAUDE.md being too long, bloated, eating context, "collapsing context", needing cleanup, or asks to set up / audit / split / restructure / generate a CLAUDE.md, even if they don't use the word "skill" or "refactor" — also use it when the user wants project memory or AGENTS.md style instructions organized for both humans and agents.

Performs thorough, expert-level pull request reviews and returns a structured report with issues categorized by severity and code snippets. Use whenever the user asks to review a PR, look at a pull request, do a code review, check a branch before merge, evaluate proposed changes, or asks "what do you think of