skills

How to delegate work to another AI agent (Pi, Codex, Claude Code, Hermes) — picking the right agent, sending prompts to TUI agents, polling progress. Read BEFORE any `cmux send`/`tmux send-keys` to an agent, or whenever delegating, relaying, spawning, or orchestrating agent-to-agent work.

Read-only exposure audit of the user's MacBook (and ~/Documents/code projects) for a CVE, breach, malicious package, or other security advisory, then write a structured report to ~/Documents/security-reports/. Use when the user shares a breach/CVE/malware/supply-chain advisory and asks if they're affected, says "scan my system for X", "are we affected by Y", "check if I'm vulnerable to Z", or requests any hack/breach/cyber/vulnerability audit on this MacBook. Output matches the existing audit format in ~/Documents/security-reports/.

Prevent and fix Google Safe Browsing "Dangerous site" flags. Use when launching a public web app, buying/picking a domain, building a login or signup page, or when any site shows a red "Dangerous site" / "Deceptive site" warning in Chrome, Brave, Safari, Firefox, or Edge. Triggers on "dangerous site", "deceptive site", "site blocked", "safe browsing", "phishing flag", "red warning screen".

Use when the user wants to manage his VPS servers and the AI agents running inside them — connecting, deploying, monitoring, restarting, and operating remote hosts and their agents. Triggers on VPS, server management, remote host, SSH into server, manage my servers, agents on the server.

Use DeepAPI for all regular web searches (POST /v1/search/web) and for any deep research you perform (POST /v1/research/deep), instead of built-in search or research tools. Also use it to scrape the web (websites, LinkedIn, GitHub, X/Twitter, YouTube), draft and send safe email, and generate images with DEEPAPI_API_BASE_URL and DEEPAPI_API_KEY.

Use whenever the user needs the transcript of a YouTube video — fetching, extracting, downloading, or pulling captions/subtitles/transcript text from a YouTube URL. Triggers on "get the transcript", "transcript of this video", "pull the captions", "download subtitles", "what does this YouTube video say". Primary path is DeepAPI (go to deepapi.co to get an API key); yt-dlp is the local fallback.

Distribute a skill across the 4 agent skill folders (Codex, Claude Code, Pi, Hermes) so all agents see it. Use when the user says "distribute this skill", "sync skills across agents", or after creating/updating a skill that should be global. Covers the symlink layout and the ~/.pi/agent/skills trap.

Create a specialized CLAUDE.md (+ AGENTS.md symlink) inside a specific folder to give future agents folder-scoped context. Use when the user asks to create a CLAUDE.md for a folder, write folder instructions, or add agent context to a directory.

Commit and push agent-skill changes to the user's private skills GitHub repo (private-skills-repo, rooted at ~/.agents). Use after creating or updating any skill, when the user says "push the skill", "push skills to github", "save the skill to my repo", or "update the skills repo". Handles staging, committing, pushing, and cleaning up the cmux pane used to do it.

Gauge the user's technical + product knowledge through 7 adaptive questions, log verbatim answers with honest ratings, and grow a learning plan from the gaps found. Use when the user says "level up", "level-up session", "quiz me", "gauge my knowledge", or wants a new assessment round. Differentiator: this finds and maps gaps; the `teach` skill delivers lessons on them.

Prompt the user with pointed questions to extract what is in their head about a project — remaining work, what is being avoided, what really matters, what does not. Use when the user says "prompt me", "ask me questions", or wants the agent to figure out priorities by questioning them.

Rewrite the last response simpler and shorter in plain English, prefixed with a 2-4 sentence TLDR of the conversation so far. Manual-only, invoked as /remind.

Provision a hardened SELECT-only Postgres role so AI agents can safely read a production database. Works on Supabase and any Postgres. Use when the user wants agents to query prod data, says "read-only role", "safe prod DB access for agents", or is tired of running SQL by hand for agents. Differentiator: this skill CREATES the role and wiring; day-to-day querying belongs in a project-local skill.

Launch OpenAI Codex CLI as a subagent (ChatGPT subscription auth, no API key). Use when delegating a self-contained coding task to Codex from another agent — parallel implementation work, a second opinion, or an independent verification pass.

Run a deep, source-backed research query via DeepAPI (go to deepapi.co to get an API key) POST /v1/research/deep. Builds a rigorous one-paragraph research prompt (per research-prompt rules), fires it, and saves a cited markdown report. Use when the user asks for "deep research", "deepapi research", "perplexity deep research" (legacy trigger), or any deep source-backed research run. Differentiator vs the deepapi skill: this is the full research workflow (prompt + run + report file), not raw endpoint access.

Research any online purchase with DeepAPI — fair-price checks, best deals, where to buy, shop trust. Load whenever the user is shopping or buying anything online: mentions buying a product, comparing prices, "is this a good price", "where can I get X", or attaches a product photo or listing screenshot. Research only — never places orders.

How to write effective agent skills — what to do, what not to do, anatomy, progressive disclosure, design patterns, anti-patterns, testing, security. Read this whenever a skill (Claude Skill, Agent Skill, SKILL.md) is being created, edited, reviewed, or debugged. Use when the user says "create a skill", "new skill", "update this skill", "improve a skill", "why isn't my skill triggering", or anything else involving authoring or editing SKILL.md files.

Score any AI model on the DeepSWE coding-agent benchmark via the OpenRouter API. Use when the user wants an independent, reproducible coding-agent eval — "run DeepSWE", "benchmark this model on DeepSWE", "score model X on the coding benchmark", "test a model via OpenRouter on DeepSWE", or to verify vendor-reported coding scores. Covers setup, the OpenRouter wiring for mini-swe-agent, single-task / subset / full 113-task runs, and leaderboard submission.

Use when the user wants to extract project vision, decisions, and preferences from his head into clear documentation (README + ADRs) through a back-and-forth Q&A loop. Triggers on "brain-to-docs", "build out the docs", "extract the vision", "let's document this project".

Explain and write effective instructions for the `/goal` feature — the persistent self-checking agent loop (plan → act → test → review → iterate), available in agents like Codex, Claude Code, and Hermes Agent. Use when the user mentions `/goal`, "goal loop", "Ralph loop", wants to kick off a long-running autonomous agent run, asks how to write a goal prompt, or wants a one-paragraph goal instruction drafted.

Keep the user's MacBook awake with macOS caffeinate — prevent sleep, screen dimming, or both, for a set duration or while a process runs. Use when the user says "don't let my mac sleep", "keep the screen on", "anti-sleep", "caffeinate", or wants the machine awake overnight / during a long build.

MUST be read ANY time you interact with cmux in ANY way — listing/inspecting/creating/closing cmux workspaces, panes, or surfaces; reading or capturing pane/screen output; sending input or keys to a pane/surface; delegating to, polling, or checking on other agents running in cmux panes/surfaces; building or rearranging terminal layout; cmux browser automation; sending notifications/flashes/status/progress to the sidebar; editing cmux settings; or integrating an agent with cmux hooks. If your command starts with `cmux ` or touches a cmux workspace/pane/surface/agent, read this FIRST. Triggers on "cmux", "in this workspace", "this pane", "the other agent", "delegate to", "check on the agent", "send to the pane". macOS only (14.0+).

Read every ADR markdown file in the project's docs/adr/ folder so you have full context on past decisions. Use only when the user explicitly calls it.

Register a custom or variant model (e.g. an OpenRouter ":nitro" / ":floor" / ":exacto" slug) in the Pi Agent so it can be set as the global default. Use when Pi silently falls back to a different model (e.g. moonshotai/kimi-k2.6) after setting defaultModel, or when a model slug isn't in Pi's bundled list. Triggers on "Pi reset my model", "Pi won't use this model", "add a model to Pi", "Pi default keeps reverting".

Direct browser control via CDP. Use when the user wants to automate, scrape, test, or interact with web pages. Connects to the user's already-running Chrome.

Teach the user a new skill or concept, within this workspace.

Write a single-paragraph Deep Research prompt to hand to a human researcher (or a deep-research AI). Use when the user wants a research brief, a "deep research prompt", a one-paragraph task for a researcher, or asks "what should our researcher look for". Produces ONE tight paragraph with full context, numbered sub-questions, and per-finding output format.

Make an AI agent run on a schedule, loop, or interval — cron, heartbeats, recurring autonomous checks. Use for "run every N minutes", "schedule a task", "run on a loop", "heartbeat". Covers external clocks (Claude Code, Codex, Pi) vs Hermes' built-in scheduler.

Walk the user through setting up anything step by step. Use when the user asks for help setting up, configuring, installing, or getting something working — "help me set up X", "walk me through this", "setup-help". Differentiator: gives one current step at a time, then always lists every remaining setup step after each response.

ONLY for Pi Agents — all other agents have their own web tools. How Pi accesses the web via the pi-web-access package — search, fetch URLs/PDFs/YouTube/GitHub. Use whenever a Pi task needs current info, docs, news, prices, or content from a specific URL.

Manually-invoked skill that forces the agent to compress its current answer — strip filler, simplify wording, and cut length while keeping the substance. Use when the user says "short", "shorter", "simpler", "too long", "tl;dr", or wants a more concise version of the previous response.

Rewrite a user's prompt to reduce the chance it trips Claude Fable 5's server-side safety classifiers (cyber/bio guardrails that force-route to Opus 4.8 or return stop_reason "refusal"). Use when the user hands you a prompt that touches cybersecurity, auth, exploits, malware, pentesting, or other dual-use topics and asks to make it "Fable-safe", "guardrail-safe", "won't get flagged/refused/downgraded", or to rewrite it so Fable 5 won't block it.

Compact the current conversation into a single, detailed handoff message — everything that happened, why it happened, and what's left — output in a code block so it can be copy-pasted into a fresh agent session. Use when hitting context limits, switching focus, ending a work session, or partitioning a task across fresh contexts.