skills

Run a long drive as three Claude Code sessions: a communicator that acks every message of the owner's within seconds and writes everything he sees, a coordinator that turns each order into a background agent shipping it end to end while one watcher turns CI, merges, taps and promised times into events, and a meta session that owns every fix to the drive's own tools, skill rewrite and install.

1 skill · 4 agents · 1 MCP server · hooks

Get a second opinion from OpenAI Codex CLI on hard debugging, analysis, or architecture problems, or generate images with its $imagegen skill.

1 skill · 2 agents · 2 MCP servers · hooks

Open a PR in the repo's own voice — guidelines and template from ccx, house style scouted from recent human-authored merged PRs, an OSS politeness and standards pass, shipped via ccx vcs ship, and a background watcher that clears mechanical CI and bot failures.

2 skills · 2 agents

Run authenticated agent-browser sessions by reusing your local browser login — cookies for one or more hosts behind one Touch ID tap (cookiesync CLI), through one ab entry point: Browserbase cloud when its key resolves, local stealth browser otherwise. macOS.

1 skill · hooks

Generate animated SVG terminal demos of a CLI with evp — write a .tape, render it (native on Linux x86_64, Docker elsewhere), inspect keyframes, and refine in a loop.

1 skill · hooks

Create or refresh a fancy GitHub profile README from your real repos and activity — AI banner, snake animation, cron Actions, and a daily Claude refresh that summarizes recent commits and releases.

2 skills

Launch Orca workers on a Sprite or a local worktree, reclaim what a settled Dispatch left behind, and diagnose a stuck lane, with a repository's values checked in once as .orca.json.

1 skill · 1 agent

Bootstrap new projects and repos with base conventions plus opinionated Python, Go, and Swift layers (SPM package/CLI or Xcode app), each with opt-in docs/release features.

2 skills

Run slop-cop on a file to catch LLM-generated prose tells.

1 skill · hooks

Run an assumptions-first architecture design — stable-ID registers, decision-by-question rounds, adversarial review, spikes — and ship it as an interactive HTML doc that opens on a summary deck that runs as one argument, draws its system card with Mermaid under a pan-and-zoom that survives the deck, cites entries by readable handles in parentheses, opens every numbers table on its claim, previews every cited source line from GitHub on hover, twins every entry with plain English, hosts declared interactive components and self-laying-out compare figures, links every decision and open item to the pull requests that land it with live GitHub state, replays the interview, and answers questions from the whole document with grounded follow-ups in a collapsed Ask bar and an inline Cmd-K palette behind a deployed key, plus a PDF printed from the same page, written as a humble proposal.

1 skill

Generate project images — mascot logos, README banners (dark/light), social cards, illustrations — via the OpenAI Images API with local under-1MiB compression.

1 skill

Write incident retrospectives as an interactive HTML page rendered from one canonical retro.json: outage windows and time-to-detect/mitigate/resolve tiles derived from timestamps, a swimlane timeline with Slack quotes, root and contributing causes with evidence, action items with live GitHub state, committed Datadog notebook and monitor snapshots drawn in the page, reader comments, and an in-page assistant; imports Google Docs postmortems.

1 skill

A live dashboard assembled per task from typed components: PRs waiting for review, stack landing previews, Slack test-thread feeds, latency percentiles, design gates, Datadog monitors, cci lanes, and local components written for the job at hand, served on a stable port with a scheduler that caches, backs off, and drops secret-shaped payloads.

1 skill

Guidance for writing effective LLM prompts and agent instructions, with current per-provider model notes.

1 skill

Write a plan in plan mode with code-grounded fix steps and a workflow table, and answer a code review that re-enters plan mode with every comment carried verbatim.

1 skill

Maintain a Claude-written summaries sidecar in any repo — a committed read-side module plus a config-driven daily refresh skill that summarizes real commit and release data into one-line suffixes.

1 skill

Choose the right surface before delivering to the human — prose, AskUserQuestion, a static Artifact page, or a live cc-present board.

1 skill

Apply Diataxis modes, a technical-builder voice, the canonical README skeleton (one live install path, install-to-quickstart coherence, no internals), and a slop-cop prose-lint pass when writing project documentation, READMEs, tutorials, how-tos, or reference.

1 skill