dotfiles

Full-lifecycle orchestration for comprehensive, no-compromise execution — roll-calls the right skills, fans out parallel/worktree agents, then closes the loop (verify, review, simplify, changelog, handoff, retro). Use when you want maximum effort, to go all-in, or to orchestrate a big task end-to-end.

Runs a short retrospective that PROPOSES edits to the user's Claude Code skills — capturing a newly-found gotcha, fixing a description that didn't auto-trigger, retiring a dead skill, or extracting a reusable pattern into a new skill. Improves the TOOLSET, not the session log. Use when the user signals satisfaction at the end of real work ("thanks", "thanks that worked", "that worked great", "nice", "perfect", "exactly what I needed"), or when they explicitly run /session-retro. Always proposes and waits for confirmation — never edits skills silently. In unattended runs (/orchestrate, overnight, or /session-retro --auto), writes proposals to ~/.claude/retro-proposals/ instead of blocking.

Generate a concise session handoff for resuming work later, including current state, decisions, open issues, and next steps.

Deep task decomposition into parallel workstreams with self-contained agent prompts. Use when you want to review the execution plan before committing to it.

Generate a handoff note for clean session transitions — preserves context across /clear or new sessions. Use when ending a session, when context is getting high, or when the user says "hand off", "wrap up the session", or "save state for next time".

Inspect and clean stale branches across multiple repositories using the dotfiles git-hygiene + gh-bootstrap toolchain. Use when the user asks about branch state, stale local branches, or "no ref was fetched" errors.

Update a project CHANGELOG.md or equivalent session log with concise notes about completed changes, decisions, and known issues.

Commit current work, push the branch, and create a GitHub pull request with a concise title, body, and verification notes.

Pick up a GitHub issue or issue-like bug report end to end: investigate, reproduce when useful, implement, verify, commit, push, and prepare a PR when requested.

Produce a read-only repository health briefing covering stack, git state, setup, tests, dependencies, blockers, and next steps.

Review recent or specified code changes for correctness, security, quality, and missing verification. Use when the user asks for a review, pre-ship check, PR review, or quality/security pass.

Review and simplify recently changed code by removing unnecessary complexity without changing behavior. Use when the user asks to simplify, reduce over-engineering, clean up a patch, or make code easier to maintain.

Read-only review of recent git changes (diff since the last review/tag) for correctness, security, and quality — flags edge cases and missing boundary error handling, hardcoded secrets, injection risks (SQL/XSS/command), sensitive data in logs, over-engineering, and dead code, then returns a prioritized checklist. Reports only; does not edit — to actually apply simplifications use simplify. Use when the user asks to "review my changes", "check this before I ship", "look over the diff", "is this safe/correct", or wants a quality/security pass on recent work.

Run an automated code review using your regular Antigravity subscription plan. Use when the user asks for a Gemini/Antigravity review, a second opinion on recent changes, or /antigravity-review. Reviews the working-tree diff for bug risks, boundary-condition errors, security issues, and over-engineering — the Gemini sibling of the Codex review gate.

Use when handed a runtime or browser verification task: prove a change actually works end-to-end in a running app, adversarially. Requires the Playwright MCP server (global mcp_config). Produces a verdict plus evidence other agents can audit.

Commit, push, and create a PR in one shot — Boris Cherny's most-used daily command

Records what changed this session into a CHANGELOG.md at the project root — reads git log since the last entry and prepends a dated entry (What changed / Decisions made / Known issues), then commits it. Use when the user explicitly wants to log/record changes — "update the changelog", "log this session", "add a changelog entry", "record what we did". For preserving context across /clear or a new session use handoff instead; for improving your skills use session-retro.

Review recently changed code for unnecessary complexity and edit it in place to simplify — removes over-engineering, dead code, and premature abstractions while preserving behavior. Unlike review (which only reports), this applies the changes. Use after building a feature, or when the user says "simplify this", "this feels over-engineered", "clean this up", or "de-engineer".

Recalibrate to the Fable conduct layer — the operating discipline distilled from Claude Fable 5 in FABLE.md. Use at the start of a session on any model, when output drifts (terse fragments, arrow chains, asking permission for reversible work, unverified claims, turns ending on promises), or when the user says "fable mode", "act like Fable", "recalibrate", "you're drifting".

Bootstrap a repo's doc contract or audit doc drift — declares every markdown surface (LIVING/GENERATED/SOURCE/HISTORICAL), wires the CI doc-truth checker, migrates shadow trackers to GitHub issues, and reconciles out-of-band drift (closed issues, migration high-water marks, stale PRs, ghost worktrees). Use when the user says "drift sweep", "bootstrap the doc contract", "are these docs stale", after a rename or process retirement, or pre-launch.

Starts a Claude Code remote-control server via `claude remote-control --spawn worktree`, running in an isolated git worktree so remote work doesn't touch local changes, and returns the connection info for claude.ai/code and the Claude mobile app. Use when the user wants to control this project remotely, asks to "start the remote server", "connect from my phone/mobile", "access this from claude.ai", or "spin up claude-server".

Pick up a GitHub issue, investigate, implement the fix, test, and create a PR

Drives jujutsu (jj) version control on the user's behalf so they don't have to memorize commands. Use for single-agent feature work (the default), whenever the user mentions jujutsu/jj, when undo-safety or a clean rewritable history matters, or when starting/finishing a change in a jj-managed repo. For MULTI-agent parallel work use git worktrees (or jj workspaces) instead — see "jj vs worktrees" below.

Scaffolds a new project in the current directory — creates CLAUDE.md, CHANGELOG.md, .gitignore, README.md, and a source dir, initializes git with a first commit, and offers to create a GitHub repo. Prompts for project name, description, and language/framework first. Use when the user wants to "start a new project", "bootstrap a repo", "set up a fresh project", "scaffold a codebase", or "kick off" a new build from scratch.

Records a recurring failure into ERRORS.md — classifies it (hallucination, instruction-ignored, context-lost, wrong-tool, incomplete, external), captures the goal, exact error, attempts tried, root cause, and lesson, then commits it; also logs wins on request. Use when the same error keeps recurring, an approach repeatedly fails, the user says "log this error", "document this failure", "this keeps happening", or "log that as a win" after a success.

Inspect and clean stale branches across multiple repositories using the dotfiles git-hygiene + gh-bootstrap toolchain. Use when the user asks about branch state, stale local branches, or "no ref was fetched" errors.