agent-browser
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction.
dogfood
Systematically explore and test a web application to find bugs, UX issues, and other problems. Use when asked to "dogfood", "QA", "exploratory test", "find issues", "bug hunt", "test this app/site/platform", or review the quality of a web application. Produces a structured report with full reproduction evidence -- step-by-step screenshots, repro videos, and detailed repro steps for every issue -- so findings can be handed directly to the responsible teams.
browser-proof-video
Capture browser proof artifacts and render final verification videos. Use when proof-driven verification needs Agent Browser screenshots/WebM/logs, artifact validation, Deepgram narration, or Remotion MP4 output.
code-review-swarm
Run a structured code-review swarm for a PR, branch, or diff. Use when proof-driven verification needs read-only review shards, especially when subagents are explicitly authorized or the diff exceeds 10 files or 1,200 diff lines.
computer-use-fallback
Use Computer Use as a fallback when proof-driven verification cannot be completed with structured browser automation. Applies to native desktop apps, browser extensions, existing logged-in browser profiles, iOS simulators, file pickers, desktop GUI-only flows, or bugs that only reproduce in a real GUI.
remotion-best-practices
Best practices for Remotion - Video creation in React
verification-setup
Prepare or diagnose the proof-driven verification environment. Use when asked to set up, install, upgrade, doctor, validate prerequisites, or troubleshoot Agent Browser, ffmpeg, Deepgram CLI, Remotion, Next.js DevTools MCP, or GTM Hub dev-server readiness for proof-driven verification.
verify-pr-proof
Single-invocation ProofOps orchestrator for verifying a PR, branch, or completed coding task end to end with definition-of-done gating, Code Review agents, Browser QA agents, Developer Fixer agents, optional framework diagnostics, Agent Browser evidence, iterative fixes, validated artifacts, optional Remotion plus Deepgram proof video, and final commit. Use when the user asks to verify, review, prove, QA, test, inspect a PR/branch/task, create evidence, produce proof of work, or run ProofOps.
qa-video-director
Use after QA has passed to plan and produce polished founder-facing proof videos as separate intro, deliberate walkthrough, and optional closeout segments with synchronized narration, visible cursor/clicks, validation, and ffmpeg stitching.
qa-video-producer
Use after QA has already passed to produce founder-facing proof videos with a Browser Use-verified workflow, a Playwright-recorded walkthrough, Remotion intro/outro, and Deepgram narration. Never use for partial, failed, or blocked QA runs.
qa
Use when the user asks Codex to set up the QA plugin pipeline, verify completed work, review code, run tests, debug Next.js or Convex behavior, use Browser Use for real browser QA, fix issues, update documentation, create narrated proof videos, explain QA/plugin behavior, commit changes, or verify one or more completed sprint folders.
sprint-execute
Use when the user asks Codex to execute a sprint, implement all sprint stories, run a multi-agent development sprint, or carry stories through code changes and commits.
sprint-research
Use when the user asks Codex to research a feature, review an existing plan/spec, size work into one or more sprints, create sprint artifacts, or turn broad requirements into a sprint distribution.
sprint-review
Use when the user asks Codex to review sprint stories, audit story quality, incorporate feedback, check story sizing, or create the verification checklist before implementation.
sprint-stories
Use when the user asks Codex to write sprint stories, convert a sprint plan into implementation stories, generate STORY files, or prepare a reviewed sprint backlog.
sprint-verify
Use when the user asks Codex to verify a completed sprint, run browser QA, produce evidence, fix verification bugs, invoke Browser Use or verification plugins, create a verification report, or prepare a PR.
audit-run
The full audit→fix cycle for Millwork OS production runs — invoke with a run/chat URL (or thread id) plus the requester's brain-dump of observations. Use whenever the user says "audit this run", pastes a millos.ai chat URL with complaints, asks why a run was slow/wrong/expensive, or says "execute the audit findings". Covers evidence collection (checkpoints, Langfuse, DB, VFS artifacts, Vercel/Railway logs), cost+time accounting, two-persona judgment, problem documentation, the fix plan with execution lanes, and — after approval — the agent-team execution, verification loop, build gate, PR, and optional live verification + deployment.
autosprint
Plan and autonomously execute engineering sprints end to end — research a problem with small agent teams, write an epic/task sprint plan with self-contained per-task specs, launch the bundled dynamic-workflow executor (one agent per task, per-task commits, autonomous epic reviews, live progress dashboard), and hand off a PR with a completion report and human verification guide. Use when the user wants to plan or run a sprint, audit a problem before sprinting it, turn research or a PRD into autonomously executable work, launch overnight/unattended implementation, check on or resume a running sprint, or hand finished AI work to engineers for verification. Successor to sprint-protocol, built on Claude Code dynamic workflows, with optional Codex CLI task routing.
essayist
Turn rough dictated thoughts into a publishable technical essay in the author's own voice — researched, fact-checked, cited, illustrated with figures, and proofread by parallel review agents before delivery. Use whenever the user wants to write or draft a technical article, essay, blog post, X/Twitter long-form post, or Medium piece; says "write this up", "turn my notes/thoughts into a post", "help me write about what i learned/built"; or dictates a messy stream of thoughts about a technical topic they want published. A raw voice-transcription braindump is the expected input, not a problem to fix.
launchpad
Decide HOW to run a coding task — interactive, goal mode, or a workflow — then author the launch-ready prompt or goal package for Claude Code, Codex, or Antigravity. Use when the user says "launchpad", "start a goal", "set up a goal/sprint", "should this be a goal or interactive", "plan an autonomous run", "I want to brain-dump a task", "help me write a goal prompt", or is about to kick off a long autonomous run and wants it scoped, delegated, and verifiable first.
prd
Create self-verifying PRDs for autonomous execution. Interviews users to gather requirements, then generates structured prd.json with phased implementation and appropriate testing strategies. Supports 7 task categories with type-specific workflows. Use when user says "create a prd", "prd for", "plan a feature", "plan this", "write prd", or wants to plan any multi-step implementation work.
sprint-protocol
Use when turning founder requirements, research packets, plans, specs, and system designs into one or more independently executable Claude/Cloud Code implementation sprints with story writing, review, execution, optional verification, and sprint-level commits.
ralph-preflight
Pre-flight check for Ralph TUI loops. Validates config, templates, prd.json, and environment before starting a loop. Run after /prd to verify everything is ready. Detects global CLAUDE.md conflicts, validates template variables, and provides launch commands.
setup-claude
Interactive Claude Code repository setup and optimization. Configures the complete ecosystem - skills, commands, subagents, hooks, rules, MCPs, and plugins. Invoke with /setup-claude init or /setup-claude audit.