Takes a step back and critiques the current plan or approach from a skeptical outside perspective before continuing, to catch flawed assumptions early. Use when the user says "/advisor", "play devil's advocate", "sanity check this approach", or before committing to a large or risky change.
Reviews the current diff or a specified target for correctness bugs, security issues, performance problems, and style, at a chosen effort level. Use when the user says "/code-review", "review my changes", "review this PR", or asks for a thorough code review before shipping.
Produces a condensed summary of the current session's key decisions, changes made, and open threads, so the user can clear the conversation and continue with minimal context loss. Use when the user says "/compact", "the context is getting long", or "summarize this session before we continue".
Runs a structured debugging workflow for a specific error or unexpected behavior — reproduces it, isolates the cause, and proposes a fix, with debug logging enabled along the way. Use when the user says "/debug", describes a bug, pastes an error/stack trace, or says something isn't working as expected.
Runs a structured multi-step research process on a topic or question, using web search across multiple angles before synthesizing a thorough answer, rather than answering from a single search. Use when the user says "/deep-research", asks for a thorough investigation, or the topic is broad enough that one search…
Diagnoses common environment and setup issues for the current project — missing dependencies, misconfigured tooling, broken scripts referenced in AGENTS.md. Use when the user says "/doctor", "something's broken with my setup", or the agent's own commands keep failing.
Exports the current session's key content (decisions, code changes, explanations) to a markdown file the user can keep or share. Use when the user says "/export", "save this conversation", or "write this session out to a file".
Scans the current project and generates or updates an AGENTS.md file containing directory structure, build/test/lint commands, coding conventions, and architecture notes. Use when the user says "/init", "initialize this project", "set up project memory", or asks the agent to learn the codebase's conventions for the…
Opens, edits, or appends to the project's AGENTS.md memory file, or records a quick standing instruction. Use when the user says "/memory", "remember that...", "add to project memory", or "always do X in this project".
Gives a one-line (or a few line) summary of what's been accomplished in the current session so far. Use when the user says "/recap", "what have we done so far", or "give me a quick summary of this session".
Gives a fast, single-pass review of a pull request or the current changes — lighter weight than /code-review, meant for a quick sanity check rather than a deep audit. Use when the user says "/review", "quick review", or "sanity check this PR".
Runs a focused security vulnerability scan on the current diff or codebase, checking for injection risks, exposed credentials, auth issues, and unsafe configuration. Use when the user says "/security-review", "check for vulnerabilities", or "audit this for security issues".
Runs a cleanup-only pass over the current diff or a target file — reducing unneeded abstraction, clearer naming, less cleverness — without hunting for bugs. Use when the user says "/simplify", "clean this up", or "make this simpler" as distinct from a full code review.
Scaffolds a new Antigravity skill (SKILL.md plus optional scripts/examples/resources folders) from a plain-language description of what it should do. Use when the user says "/skill-creator", "make a new skill for X", "turn this into a skill", or wants to package a repeated workflow as a reusable command.
Generates a teammate ramp-up guide by inspecting the project's AGENTS.md, installed skills, recent git history, and structure, then producing an onboarding document. Use when the user says "/team-onboarding", "write an onboarding doc", or "help a new dev get up to speed on this project".
Gives a best-effort summary of the current session's activity (tool calls made, files touched, approximate turn count) as a stand-in for Claude Code's /cost token/usage tracking. Use when the user says "/usage-report", "/cost", "how much have we used this session", or asks about session cost.
Checks that a completed implementation actually satisfies the original request or spec — running tests, checking edge cases, and confirming nothing was missed — before declaring a task done. Use when the user says "/verify", "double check this works", or before marking a task complete.