ouroboros

Detect regular rectangular board cells and click untried non-HUD cells while avoiding prior no-op, HUD-only, or death outcomes.

Prefer safe untried simple actions from the current state when no higher-confidence solver has a plan.

Use inferred direction deltas to search toward unvisited safe positions after simple action probes identify a controlled object.

When coordinate clicks are available but no board model is confident, probe visible compact non-HUD objects before raw fallback actions.

Play ARC-AGI-3 interactive grid games via the arc MCP tools — explore mechanics from observation, then exploit them to win levels.

Core agent-browser usage guide. Read this before running any agent-browser commands. Covers the snapshot-and-ref workflow, navigating pages, interacting with elements (click, fill, type, select), extracting text and data, taking screenshots, managing tabs, handling forms and auth, waiting for content, running multiple browser sessions in parallel, and troubleshooting common failures. Use when the user asks to interact with a website, fill a form, click something, extract data, take a screenshot, log into a site, test a web app, or automate any browser task.

Use this skill when the user wants Ouroboros to browse or automate websites in a real Chrome-compatible browser. Prefer it for navigating pages, taking browser snapshots, clicking or filling visible controls, scraping page state, testing web flows, and using CDP auto-connect without asking the user to install npm packages.

Run Ouroboros desktop intent-based E2E test plans written in Markdown by driving the Electron app through Agent Browser over Chrome DevTools Protocol. Use when asked to execute, debug, or author natural-language desktop QA charters in test-plan/.

Create intent-based Ouroboros desktop E2E Markdown test plans from a high-level goal, feature, module, bug, PRD, ticket, or UI area. Use when asked to generate, draft, author, improve, or expand files under test-plan/ for Agent Browser desktop testing.

Automates browser interactions via CLI using agent-browser by Vercel Labs. Covers navigation, clicking, form filling, snapshots, refs-based selectors, agent mode with JSON output, session management, and CDP integration. Use when the user needs to automate web browsing, scrape pages, fill forms, or integrate browser automation into AI agent workflows.

Answer questions about the AI SDK and help build AI-powered features. Use when developers: (1) Ask about AI SDK functions like generateText, streamText, ToolLoopAgent, embed, or tools, (2) Want to build AI agents, chatbots, RAG systems, or text generation features, (3) Have questions about AI providers (OpenAI, Anthropic, Google, etc.), streaming, tool calling, structured output, or embeddings, (4) Use React hooks like useChat or useCompletion. Triggers on: "AI SDK", "Vercel AI SDK", "generateText", "streamText", "add AI to my app", "build an agent", "tool calling", "structured output", "useChat".

Audit implemented Ouroboros features against requirements, PRDs, tickets, GitHub issues, or acceptance criteria to find missing, partial, incorrect, or unverifiable work. Use when the user asks to audit a feature, verify implementation completeness, compare code to requirements, check acceptance criteria, run a gap analysis, determine what is left to build, check whether a ticket is done, audit against a PRD/spec, or asks whether recent work in the Ouroboros codebase matches its requirements. Do NOT use for creating PRDs, generating tickets, implementing tickets, broad full-codebase review, or general bug triage.

Perform a comprehensive full-codebase code review using specialized parallel agents. Use this skill whenever the user asks for a "full code review", "codebase review", "review everything", "audit the code", "code health check", or any request to review the entire project (not just a PR or single file). Also trigger when the user says things like "what's wrong with this codebase", "find all the issues", "review all my code", or "comprehensive review". This is specifically for whole-codebase analysis, not single-file or PR-scoped reviews.

Use this skill whenever the user wants to generate implementation tickets from an existing PRD, break a PRD into engineering tickets, create ticket files from a product spec, or turn a requirements document into actionable work items. Triggers include: 'create tickets from PRD', 'break this into tickets', 'generate tickets', 'write tickets for this PRD', 'turn this PRD into tickets', 'make implementation tickets', 'ticket this up', or any request to take an existing PRD or product spec and produce individual implementation tickets. The PRD can be a local file (markdown). Also use when the user has just finished writing a PRD and wants to generate tickets from it. Use this skill even if the user just says 'create the tickets' or 'ticket this' when a PRD file is referenced. Do NOT use for writing PRDs from scratch.

Orchestrate a parallel agent team (Sam, Tim, Jack) to implement tickets from the tickets/ folder. Reads all ticket files, resolves dependencies via topological sort into execution waves, presents the full execution plan for user approval, then spawns three Opus 4.6 1M max effort teammates in parallel to implement. Each teammate works in an isolated git worktree, requires plan approval before coding, and reports progress in a structured format. Use this skill whenever the user wants to: execute tickets with an agent team, implement tickets in parallel, run tickets from a tickets folder, dispatch agents to work on tickets, or says things like 'implement the tickets', 'execute the tickets', 'run the ticket team', 'start the team on tickets', or 'work through the tickets'. Also triggers when the user references a tickets/ folder and wants parallel implementation.

Use this skill when the user needs structured planning, complex analysis, difficult tradeoff decisions, or troubleshooting under uncertainty and the agent should apply a compact meta-thinking loop with task classification, a meta-mode gate, plan extraction, critique, and concise trust signals. Prefer it when surfacing confidence, limitations, and context gaps will materially improve the answer. Do not use it for simple factual or one-step requests where extra framing would add friction without improving the result.

Refactor an AGENTS.md or CLAUDE.md file using the Progressive Disclosure Principle so that only essential, universal instructions stay at root level while domain-specific guidance moves to subsidiary files loaded just-in-time. Use this skill whenever the user wants to restructure, slim down, optimize, or declutter their AGENTS.md or CLAUDE.md file, mentions "progressive disclosure" for agent instructions, says their instruction file is "too long" or "bloated," asks how to reduce token overhead in their agent config, or wants Claude to "only load what it needs." Also trigger when the user says things like "refactor my CLAUDE.md," "split up my AGENTS.md," "organize my agent instructions," or "my AGENTS.md has gotten out of hand."

Perform a comprehensive full-codebase code review using specialized parallel agents. Use this skill whenever the user asks for a "full code review", "codebase review", "review everything", "audit the code", "code health check", or any request to review the entire project (not just a PR or single file). Also trigger when the user says things like "what's wrong with this codebase", "find all the issues", "review all my code", or "comprehensive review". This is specifically for whole-codebase analysis, not single-file or PR-scoped reviews.

Orchestrate a parallel agent team (Sam, Tim, Jack) to implement tickets from the tickets/ folder. Reads all ticket files, resolves dependencies via topological sort into execution waves, presents the full execution plan for user approval, then spawns three Opus 4.6 1M max effort teammates in parallel to implement. Each teammate works in an isolated git worktree, requires plan approval before coding, and reports progress in a structured format. Use this skill whenever the user wants to: execute tickets with an agent team, implement tickets in parallel, run tickets from a tickets folder, dispatch agents to work on tickets, or says things like 'implement the tickets', 'execute the tickets', 'run the ticket team', 'start the team on tickets', or 'work through the tickets'. Also triggers when the user references a tickets/ folder and wants parallel implementation.