Skills
97,222 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Intercept the AG2 agent loop with `BaseMiddleware` — wrap full turns (`on_turn`), each LLM call (`on_llm_call`), each tool execution (`on_tool_execution`), or each human-input request (`on_human_input`). Use for retry, logging, history trimming, request mutation, tool auditing, guardrails, or rate limiting. Built-ins: `LoggingMiddleware`, `RetryMiddleware`, `HistoryLimiter`, `TokenLimiter`, `TelemetryMiddleware` (see `ag2-telemetry`). For per-tool hooks see also `ag2-add-custom-tool` tool-middleware section.
Open an AG2 network `discussion` channel — N-party (2+) round-robin where each participant speaks in fixed order, cycling until explicit close or TTL. Use when the user wants a brainstorm with a fixed cast, a panel discussion, or round-robin reviewers. Covers `agent_client.open(type="discussion", target=[...], knobs={"ordering": ORDERING_ROUND_ROBIN})`, the `expected_next_speaker` rotation, the `hc.can_send(...)` probe pattern (default handlers skip LLM calls when it isn't their turn), `DiscussionState`, the `turn_within` expectation defaults (`warn` at 120s / `hide` at 600s), view-window sizing for N participants, and the four close patterns that work with this adapter. Load this after `ag2-network-quickstart`. For conditional handoffs or declarative orchestration, see `ag2-network-workflow` instead.
Creates structured agent definitions using the 7-component format grounded in persona science (the alignment-accuracy tradeoff), vocabulary routing, and the MAST failure taxonomy + Forge watchlist. Produces agents with real-world job titles, expert domain vocabulary payloads (15-30 terms), explicit deliverables, decision boundaries, imperative SOPs, and named anti-pattern watchlists. Use this skill when the user wants to create an agent, define a role, build a persona, or needs a specialized AI assistant for a specific domain. Also triggers when Mission Planner delegates agent creation for team roles. Works for any domain — software, marketing, security, operations, design, writing, research, and more. Do NOT use for creating skills (use Skill Creator) or team composition (use Mission Planner).
Use this skill when you receive a 402 Payment Required response that contains an `agentkit` extension. Covers checking 402 responses for the AgentKit extension before paying, constructing and signing a CAIP-122 challenge (SIWE for EVM), sending the signed `agentkit` HTTP header, and interpreting access modes (free, free-trial, discount). Supports both EOA wallets (EIP-191) and Smart Contract Wallets (ERC-1271, e.g. Coinbase Smart Wallet, Safe).
Plan and deliver repository-aware AI-assisted software work through Agile Iteration Method using intent discovery, PO-owned Epics, TDO-owned User Stories, cross-Epic delivery planning, explicit gates, review, validation, and user-owned acceptance. Use for discussing product direction with repository context, creating and refining Epics and User Stories, planning delivery selections, implementing work, reviewing delivery, configuring project specialists, calibrating repositories and sharing local learning, consolidating knowledge across local AIM projects, controlling AIM modes and cost profiles, opening and controlling local AIM UI, and continuing AIM runs
Split a completed PostHog task run into activity records — what the agent tried, whether it worked, what blocked it — and record each one through the report_activity tool. Use when a task asks to analyze a run, produce a task analysis, or review a run from an attached run log. Covers the log query protocol (bounded jq queries over the raw JSONL), both log schemas, the activity schema, and evidence verification. Records facts only; it does not suggest fixes.
Run a rigorous autonomous experiment loop for any optimization target using explicit hypotheses, repeated trials, structured experiment logs, and local HTML reports. Use when asked to "run autoresearch", "optimize X in a loop", "start experiments", or "improve this with benchmark-driven iteration".
Build a layered, LLM-friendly documentation system for a project. For existing projects: scan the project structure → classify the architecture → generate the CLAUDE.md main index → base module documentation → business module API/data model/pitfall documentation → configuration documentation → git changelog (with risk assessment + rollback guide) → cross-validation. For new projects: parse the PRD → design the architecture according to scaffolding standards → break requirements down into development tasks → merge with code standards to generate the CLAUDE.md development guide → module-level API design + data models + development checklist. Use when the user needs to create AI programming documentation for a project, establish a project index, sort out code, plan a new project, or mentions llm.txt/CLAUDE.md.
Side-route skill for explicit backtracking when an upstream artifact becomes untrustworthy. Use when research, PRD, or slice decomposition has been invalidated by new information and you need to clean up stale artifacts before re-running an earlier pipeline step. Not for forward progress, fresh shaping, or routine bug fixes.
Comprehensive overview of ALL Coval platform resources, their hierarchy, relationships, API endpoints, and ID formats. Use when user asks about Coval resources, data model, how things relate, what endpoints exist, or needs context about the platform structure before making API calls.
Alias for `/origin:handoff` — symmetric brief/debrief naming. Same behavior: end-of-session ritual that writes session log + project status + granular MCP captures. Invoked as `/debrief`. Use when the user prefers the brief/debrief pair over brief/handoff.
Execute AI-generated code in secure isolated E2B cloud sandboxes with advanced monitoring, MCP gateway integration (200+ tools), and multi-language support. Use for running Python/JavaScript/TypeScript/Bash code, managing sandbox lifecycle (create, pause, resume, kill, list, connect), monitoring metrics (CPU, memory, disk), handling lifecycle events, uploading/downloading files, streaming command output, environment variable management, and integrating LLMs with code execution capabilities.
Find where an AI/LLM application is failing in production and surface the failure patterns, working from real traces. Use when someone wants to understand what's going wrong with an AI feature, find and categorize failure modes, triage errors, or investigate quality issues (wrong answers, ignored instructions, hallucinations, tool misuse) — "what's failing in my agent", "surface error patterns", "why are the responses bad", "find the common failure modes", "what should I fix next". Covers scoping to one use case, finding failing traces by whichever signal fits the context (code errors, metric outliers, trace-type slices, manual review, existing-eval spikes, clustering), and reading them into a ranked failure taxonomy.
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.