dotfiles

Michael's operating + emotional coach. Operational mode — daily startup/shutdown, weekly review, pomodoro, inbox capture, daily notes (auto-loads in ~/ws/notes). Emotional/decision mode (Joe Hudson style) — use on "coach me"/"joe coach", stuck/looping/overthinking, harsh self- or other-judgment, a binary either/or decision that won't resolve, or fear, shame, loneliness, anxiety, burnout, or grief, when Michael wants to be met in a feeling rather than handed advice. Not for clinical crises (refer out).

Use when creating or processing Todoist tasks, triaging inbox items, doing daily task review, calibrating Todoist triage behavior, or turning corrections into reusable preferences. Routes to operations (CLI actions) vs calibrated triage (policy, context recovery, preference memory, evals). Trigger this whenever the user asks what to do with Todoist items, wants better task triage, or is refining how Todoist decisions should work.

Run real Deep Research across ChatGPT, Claude, and Gemini in parallel via the user's own logged-in browser (Chrome extension, zero API cost), save each original report to Notion, then synthesize. Use whenever the user wants a "deep dive", "deep research", a thorough multi-source investigation, or to research a topic across the models and compare what each finds. Drives the paid subscription products, NOT the API. NOT for single-fact lookups or ordinary web search — use web-search for those.

Use when setting up, auditing, or improving AI agent infrastructure in a repo — AGENTS.md/CLAUDE.md files, linters, architectural constraints, feedback loops, context tiering, agent specialization, or entropy management. Also triggers on "harness engineering", "agent-friendly repo", "make my repo work well with coding agents", "set up my repo for agents", or "why is my agent struggling".

Interactive Todoist triage with preference learning. Use when the user says "triage", "process my inbox", "clean up tasks", "triage my todoist", "file these captures", or mentions inbox zero. Also use when the user has a batch of raw items (voice notes, links, ideas) that need classifying and routing to Todoist projects or Obsidian. Runs an interactive confirm/correct loop that learns your routing preferences over time.

Use when the user asks to save session context, identify the current session or thread, create a resumable handoff, or prepare a Todoist/note summary that must include the working directory and session id. Works across Codex and Claude Code by detecting runtime-specific session identifiers and normalizing them into one summary.

Persistent interactive Playwright sessions for fast iterative UI debugging — js_repl lane in Codex, tmux-REPL lane in Claude Code. Use whenever driving a browser incrementally beats edit-script-rerun loops, and ALWAYS consult the no-foreground policy before any Playwright launch.

Create or update a reproducible local Docker test environment for an existing codebase, then launch and verify it with real HTTP / user-facing checks. Use when the user wants to dockerize a repo, replace managed dev infrastructure with local containers, stand up a preview stack, or prove a codebase actually runs locally in Docker.

Use when generating or editing images. Primary tool: bundled `scripts/image_gen.py` (OpenAI GPT Image 2). For non-OpenAI providers (Gemini Pro, Gemini, Imagen), use the `generate-image` CLI. Covers generate, edit/inpaint/mask, background removal, product shots, concept art, covers, and batch variants.

Use for improving and reviewing prose — cover letters, essays, reports, grant applications, speeches, emails, or any writing. Two modes — (1) iterate: refine a draft through parallel dual-reviewer passes until it stabilizes ("improve this", "make this better", "polish this writing", "keep refining until it's ready"); (2) diff-review: compare two versions and generate an interactive HTML diff with editorial commentary ("show me what changed", "create a diff", "compare versions"). NOT for code review (use git diff / code-review) or prompt/agent-instruction tuning (use prompt-engineering).

Use when developing, aligning, calibrating, or improving ANY prompt, read-trigger description, agent instruction, or output style. Two modes — (1) alignment: fit a prompt to example input→output pairs and test generalization; (2) self-improvement: turn user corrections into a growing test suite and iterate via isolated subagents. Trigger on "align this prompt", "the prompt gave wrong output", "improve/calibrate this prompt", "make this prompt self-improving", "tune the trigger description", or running a prompt improvement cycle. NOT for general copywriting or prose editing — use the writing skill for that.

Browser automation CLI for AI agents — navigate pages, fill forms, click, screenshot, scrape data, and test web apps. Use whenever the user wants to drive a website programmatically, automate any browser task, log into a site, or run a browser in the cloud (Kernel, Browserbase, Browserless). Also covers human-in-the-loop "live view" takeover, where the agent drives and a person steps in from their phone for logins, CAPTCHAs, or MFA, then the agent resumes.

Thorough automated PR review — finds bugs, checks test coverage (unit/integration/e2e/UI), writes missing tests, screenshots UI changes, fixes issues, and generates a markdown report. Use when a PR needs comprehensive review ("review this PR", "check this PR", "review the changes"), in CI/CD, or when the user wants to verify test coverage or missing e2e/UI tests. Works for Python and TypeScript repos.

Semantic search over past conversations, notes, skills, and knowledge atoms via agent-recall. Use when prior context would help — strategic or architectural decisions, complex problems, or phrasings like "last time", "we discussed", "do you remember", "how did we" — and when stuck after initial investigation, or when a problem feels like it has prior art. Do not use it for things visible in the current conversation or answerable by reading the codebase.

The canonical guide for Claude Code agent swarms (TeamCreate-based multi-agent coordination). Use when the user says "swarm", "agent team", "orchestrate agents", or when 3+ independent tasks should run in parallel with inter-agent messaging and coordination. Covers the full swarm lifecycle: planning tasks, spawning teammates, inter-agent communication via SendMessage, monitoring, verification, and cleanup. NOT for fire-and-forget subagents (use superpowers:dispatching-parallel-agents instead).

Manage runtime-specific skill profiles without deleting the canonical skill inventory. Use when the user wants to slim down Codex or Claude, switch one runtime while leaving the other unchanged, create or edit named profiles under `agents/skill-profiles`, inspect what a profile enables, or restore the original skill links after experimentation.

Send a message to Michael's OpenClaw / "puffin" agent (the WhatsApp/Telegram bot running on EC2) and get its reply back. Use whenever Michael wants to talk to, ping, test, ask, or relay something to his openclaw/puffin/clawdbot agent, or check whether the agent is responding — e.g. "ask my openclaw bot to…", "ping puffin", "is the agent replying?", "have my bot draft…". Also use to message a specific persona agent (openclaw/michael, openclaw/art, etc.).

Use when creating, modifying, or managing git worktrees, working on parallel feature branches, or isolating work from the main repository.

This skill should be used when the user asks to "create a CLI tool", "build a command line app", "add a dev command", "write a CLI script", or when editing files in ~/.dotfiles/bin/. Provides guidance on the preferred stack for building CLI tools.

Use when adding or updating environment variables or secrets. Covers global secrets (~/.env.zsh), project-specific secrets (.env.local), and documentation patterns (.env.example).

Use when user says 'remember this', when deciding where to store learnings or context, when about to edit CLAUDE.md or create a skill, or when routing any persistent information. Determines the correct storage destination among CLAUDE.md, auto-memory, rules, skills, and local overrides.

Log, analyze, and prevent recurring mistakes. Use when: (1) a mistake or near-miss occurs (self-detected or user-detected), (2) session start/end review of recent patterns, (3) user asks to review or analyze mistakes, (4) after any corrective action to verify the fix was logged. Covers logging JSONL events, running RCA, promoting/demoting guardrails, and testing prevention.

Use when about to kill a process on a port, when a port is busy and you need to free it, or when suspecting a dev server is still running.

Use when creating or switching to a git worktree inside tmux while keeping session continuity. Supports codex and claude by auto-detecting the current session and launching resume, fork, or new sessions in a split pane.

Process raw Todoist inbox items into a structured knowledge system using GTD (actions → Todoist) and PARA (reference → Obsidian). Use when batch-processing inbox captures, triaging voice notes and links, filing raw ideas into Obsidian with MOC suggestions, or running an inbox-to-knowledge-system pipeline. Handles enrichment (URL fetching, voice interpretation), clustering related items, generating Obsidian reference notes, filing Todoist actions, and producing a triage report. Self-improving via prompt-self-improvement protocol.

Process raw Todoist inbox items into a structured knowledge system using GTD (actions → Todoist) and PARA (reference → Obsidian). Use when batch-processing inbox captures, triaging voice notes and links, filing raw ideas into Obsidian with MOC suggestions, or running an inbox-to-knowledge-system pipeline. Handles enrichment (URL fetching, voice interpretation), clustering related items, generating Obsidian reference notes, filing Todoist actions, and producing a triage report. Self-improving via prompt-self-improvement protocol.

Process raw Todoist inbox items into a structured knowledge system using GTD (actions → Todoist) and PARA (reference → Obsidian). Use when batch-processing inbox captures, triaging voice notes and links, filing raw ideas into Obsidian with MOC suggestions, or running an inbox-to-knowledge-system pipeline. Handles enrichment (URL fetching, voice interpretation), clustering related items, generating Obsidian reference notes, filing Todoist actions, and producing a triage report. Self-improving via prompt-self-improvement protocol.

Use when user says "remember this", when deciding where to store learnings or documentation, when about to edit CLAUDE.md, or when creating new plugin content (skills, commands, agents). Routes info to CLAUDE.md, skills, commands, or logs.

Use when user says 'remember this', when deciding where to store learnings or context, when about to edit CLAUDE.md or create a skill, or when routing any persistent information. Determines the correct storage destination among CLAUDE.md, auto-memory, rules, skills, and local overrides.

Use when managing remote dev server EC2 instances — starting, stopping, connecting, creating new instances, or troubleshooting the `dev` CLI. Also use when the user mentions "dev server", "dev instance", "remote development", or asks to spin up a cloud dev environment.

Use when drafting email replies for Michael, generating outbound emails, or applying his email communication patterns and style.

End-to-end define-to-ship delivery loop with DAG orchestration and stage gates. Use when the user wants to turn high-level product goals into detailed stories (risks, decisions, acceptance checks), execute with subagents, run independent dual reviews, and progress tasks only when explicit gate criteria pass.

Speeds up pi_agent_rust development and verification workflows. Use when editing providers, tools, sessions, extensions, installer/uninstaller logic, or triaging regressions in this repo.

Use when recording the screen on macOS via CLI, creating screen recordings, or capturing video of a browser/app. Covers ffmpeg AVFoundation setup, recommended settings for web demos, and file size estimates. Use this skill whenever the user mentions screen recording, video capture, recording a demo, or capturing their screen — even if they don't mention ffmpeg.

Use when viewing images in the terminal with chafa. ALWAYS open images in a separate tmux pane to avoid cluttering the conversation. Handles multi-image sequences by persisting the pane ID to a temp file.

Use when user shares a raw idea ("I have an idea...", "What if we...", "I'm thinking about..."). Captures verbatim to Obsidian, scans for related notes, generates Obsidian URL, commits to git.

Use when filing notes to Obsidian vault (~/ws/notes/), organizing tasks across systems, processing inbox items, or deciding where content belongs in the PARA structure (Todoist, Notion, Obsidian). Covers vault area conventions, document maturity (idea → concept → PRD), and system ID mappings.

Use when creating or relocating project documentation files for human readers (guides, implementation notes, reference pages); not for AI agent memory or skill files. For agent knowledge routing, see memory-placement.

Teleport the current Claude Code session to the Everything Agent cloud for autonomous continuation. Use when a task will take a long time, the user wants to walk away, or they say "teleport this".

Use when searching for domain names, checking availability, or brainstorming domain ideas. Uses tldx CLI for fast bulk checks.