Skills
97,222 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Discover and install skills from Hermes, ClawHub, GitHub, and other registries. Load this skill whenever a user asks for a capability you don't already have — image generation, social media, email, calendar, finance, DevOps, search, browser automation, etc.
Expose an AG2 `Agent` over the AG-UI protocol so a frontend (CopilotKit, custom React/Next.js, or any AG-UI client) can stream responses (text and reasoning), render tool calls, sync shared state, and track sub-task steps. Wraps the agent with `AGUIStream(agent)` and mounts it in FastAPI via `stream.dispatch(...)` or `stream.build_asgi()`. Use when the user wants a web frontend in front of an AG2 agent rather than a CLI / script.
Map of AG2 capabilities and which sibling skill to reach for. Load first when the user mentions building with AG2 (ag2) but the specific feature isn't yet clear — agents, tools, model config, delegation, memory, observers, structured output, HITL, AG-UI, MCP server hosting, A2A protocol, realtime voice, telemetry, testing, or evaluation.
Runs Denario (AstroPilot-AI), a multiagent AI system for scientific research assistance that automates end-to-end research workflows from a described dataset through idea, methodology, computational results, and a publication-ready LaTeX paper. Built on AG2 + LangGraph with a cmbagent analysis backend. Use when driving the Denario pipeline (Denario.get_idea/get_method/get_results/get_paper), generating research ideas from a dataset description, auto-developing methodology, executing analysis agents, or emitting a journal-formatted (APS/AAS/JHEP/ICML/NeurIPS/PASJ) LaTeX manuscript. Part of the AlterLab Academic Skills suite.
Orchestrates the full academic research pipeline (research, write, integrity check, review, revise, re-review, re-revise, final integrity check, finalize), coordinating alterlab-deep-research, alterlab-paper-writer, and alterlab-paper-reviewer into a seamless 10-stage workflow with mandatory integrity verification, two-stage peer review, and reproducible quality gates. Use when the request mentions academic pipeline, research to paper, full paper workflow, paper pipeline, end-to-end paper, research-to-publication, or complete paper workflow. Part of the AlterLab Academic Skills suite.
Unified entry skill for awiki-cli, providing agent identity and IM capabilities, including direct messaging, group chat, and attachment send/receive; future support for end-to-end encrypted communication, and responsible for routing related tasks, minimal loading, safety rules, and confirmation rules.
Author production-grade prompts for Bolna outbound and inbound voice agents (Hindi/English/Hinglish), and diagnose-and-fix existing prompts from call transcripts. Use this skill whenever the user describes a calling-bot or voice-agent use case (e.g. lead qualification, recruitment screening, appointment booking, onboarding calls, renewal/collection calls, NPS, surveys), pastes a Bolna or voice-agent transcript and asks for improvements, asks for an Indian-language calling script, or mentions Bolna, GPT-4.1 mini voice agents, IVR scripts, telecaller bots, or similar — even if they do not name "Bolna" explicitly. This skill produces output in a very specific structure (numbered sections, Hindi-first/English-second scripted lines, paragraph-prose instructions, YAML FAQs, JSON context blocks) and that structure is non-negotiable, so always consult this skill before drafting any voice-agent prompt.
Use when creating or developing, before writing code or implementation plans - refines rough ideas into fully-formed designs through collaborative questioning, alternative exploration, and incremental validation. Don't use during clear 'mechanical' processes
Ultra-compressed communication mode. Cuts token usage ~75% by dropping filler, articles, and pleasantries while keeping full technical accuracy. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman.
Build or refactor agent-facing CLI tools with non-interactive commands, stable --help and --json contracts, idempotent operations, and --dry-run previews. Use for CLI design, agent-friction refactors, output/exit-code debugging, or automation safety. Do not use for GUI/TUI design, conversational tools, MCP server design, or general API architecture; use api-design-and-evolution for the latter.
Reduce avoidable Codex token and context usage in long-running or repository-scale work by minimizing repeated reads, inefficient polling, context drift, and unnecessary high-capability model use. Apply when users ask to save Codex quota, streamline a long task, or make an agent workflow more context-efficient; do not apply to ordinary short tasks that need no optimization.
Launch new CLI coding agents through Corral Design 1, or correctly reopen existing agent sessions inside a Corral-active Herdr repository. Use when the user asks to launch, spawn, batch-start, reopen, relaunch, or resume agents. New launches create Corral-owned tasks; resumptions reuse existing worktrees and sessions.
Create, list, inspect, and delete Bolna knowledge bases from PDF files or URLs, including multilingual retrieval settings, chunking, overlap, similarity top k, processing status, rag_id and vector_id usage. Use when a voice agent needs RAG over FAQs, policies, product docs, or web pages.
Diagnose Bolna voice-agent issues: robotic / laggy responses, agent interrupting users or never interrupting, long silences, premature hangups, webhook misses, wrong caller context, batch failures, SIP/SRTP no-audio, agent restricted by content policy, balance-low queueing, and provider connection latency. Maps symptoms to the exact field on the agent (`task_config`, `endpointing`, `incremental_delay`, `number_of_words_for_interruption`), and uses `latency_data` from `GET /executions/{id}` + raw logs to pinpoint which pipeline component (transcriber, LLM, synthesizer) is slow.
Use when writing or reviewing JavaScript/TypeScript in this repo that calls Deepgram Conversational STT v2 / Flux (`/v2/listen`) for turn-aware streaming transcription. Covers `client.listen.v2.createConnection()` / `connect()`, Flux models, and turn events like `TurnInfo`. Use `deepgram-js-speech-to-text` for standard v1 ASR and `deepgram-js-voice-agent` for full-duplex assistants. Triggers include "flux", "v2 listen", "conversational STT", "turn detection", "end of turn", "EOT", and "listen.v2".
Turn a raw idea or existing feature into a designed product SPEC by completing entity, integration, role, and expectation closure. Upserts never destroy recorded decisions. Triggers: "design-feature", "design this feature", "define product scope".
During development, save a very brief clue when encountering a non-obvious failure, user correction, or repetitive operation that could be automated. Don’t interrupt the main task or search for evidence; don’t write anything to disk for pure consultation or read-only tasks. Leave deeper investigation to agi-mode retrospectives.
Snapshots/restores Ghostty Claude Code/Codex tabs across reboots. Use before Mac updates (保存终端会话/重启前备份终端), after restart to reopen original session IDs (恢复之前的窗口/restore my ghostty tabs), audit survivors, or periodic change-only snapshots (定期增量快照/没变化不记录). Also for unmanaged/legacy Codex account switches (额度耗尽换号/换号后恢复全部 Codex); Minions native_runtime uses Account Manager. Not for --resume, screenshots or tmux.
Use when a new Godot system or feature has open design decisions — interrogates them in batched rounds, scope first, each question with a recommended answer, and records the answers before any design or code. Triggers on "grill me", "ask me first", "question me on the design", "what do I need to decide before building". Not for choosing between nodes or APIs, and not for bug fixes.
Use ChatGPT Web's GPT 5.6 Sol Pro as a verified second-opinion partner for difficult planning, architecture, debugging, business, product, content-strategy, risk-review, and Skill-design work. Use when the user asks for GPT 5.6 Sol Pro, ChatGPT Pro, a deeper outside judgment, or a file-grounded review. Default to the Codex Chrome plugin for text, model selection, file uploads, waiting, and extraction. Use OpenCLI only when the user explicitly requests it or the Chrome plugin is unavailable and OpenCLI passes preflight.
Comprehensive SAP Joule CLI (formerly sapdas CLI) assistant for managing digital assistants from the command line — compiling capabilities, deploying assistants, running BDD tests, linting, and troubleshooting errors. Use this skill whenever the user mentions "joule cli", "sapdas", "joule compile", "joule deploy", "joule test", "joule login", "joule lint", digital assistant deployment, capability compilation, DAAR files, RTA artifacts, or any task involving the Joule command line interface — even if they just say something like "deploy my assistant" or "how do I log in to Joule from the terminal". Also trigger when the user asks about testing Joule capabilities with Cucumber, linking AI assistants, managing deployed assistants, or automating Joule workflows in CI/CD pipelines.
External-research routing for confirmed code problems: triage whether the root cause is internal / external / hybrid, run a known-issue quick search before deep root-causing (platform silent failures, nested host runtimes, no code-level suspects), and evaluate industry-wide hard limits. Delegates all WebSearch discipline to effective-web-research. Referenced by PDCA hosts via frontmatter dependencies; load during the technical-analysis stage. Do NOT use for competitor product-behavior, interaction-boundary, or mechanism observation (competitive-research).
List, fetch, patch, fully update, delete, and stop queued calls for Bolna v2 voice agents. Use when the user wants to inspect agent configuration, change prompts or voices, back up an agent, decommission an agent, or cancel all queued calls for one agent.
(NS) Transform atomic implementation work into rich task markdown files for backend and frontend layers with dependencies, estimates, and layer-specific checklists. Use during SDD planning when generating task-NNN-*.md from requirements features — not for E2E or unit test tasks (use specialized generators). Requires requirements.md. Output English task files under version tasks folder.
(NS) Partition a large master requirements.md into executable subversions with version-roadmap.md and slice requirements excerpts — without generating tasks. Use when a version has too many features for one planning cycle or the user asks to split a version — after Gate 1 requirements confirmed. Do NOT generate task-NNN files (planning per slice comes later).
Use the public `orca` CLI to operate Orca-managed worktrees, folder contexts, terminals, repos, automations, worktree comments, and the browser embedded inside the Orca app. Use when the user says "$orca-cli", "use orca cli", "Orca worktree", "child worktree", "cardStatus", "spawn codex/claude in a worktree", "read/wait/send Orca terminal", "terminal send", "full handoff", "handover", "give this to another agent", "another worktree", "Orca browser", or "control the browser inside Orca". Prefer this over raw `git worktree`, ad hoc PTYs, Playwright, or Computer Use when the task touches Orca-managed state. Use Computer Use for browser windows, webviews, or desktop UI outside Orca's embedded browser.
Fetch tasks from PREQSTATION, execute work, and push updates/results back to PREQSTATION APIs. Use when asked to run a PREQ task, fetch todo tasks, or update execution status. Prefer the remote MCP endpoint at /mcp with OAuth login. Shell helper mode still uses PREQSTATION_API_URL and PREQSTATION_TOKEN. Tasks carry an `engine` field (`claude-code` | `codex` | `gemini-cli`) indicating which AI agent should execute them.
Transform AI agents from task-followers into proactive partners that anticipate needs and continuously improve. Now with WAL Protocol, Working Buffer for context survival, Compaction Recovery, and battle-tested security patterns. Part of the Hal Stack
Access and ground work in a separate Project Brain knowledge repository through the authenticated GitHub CLI. Use at the start of substantive work in any Git repository to check its root for `.project-brain.json`; when the marker exists, using this skill and consulting the brain is mandatory before planning, implementation, diagnosis, review, documentation, or project answers. Also use when explicitly asked for Project Brain context, decisions, requirements, terminology, source-system records, or knowledge-graph output.
Maintain a project's canonical terms and settled decisions that future work should reuse, and record follow-ups a session discovers but does not resolve. Use when project-specific terms are being clarified, when a choice that may constrain future work is being considered or settles, including during planning, or when a session applies a temporary workaround whose root cause stays open or observes an out-of-scope defect with evidence. Do not use for simple definitions, routine implementation details, or carrying out an already-settled decision.
Run the reviewer's half of the document review loop without the user relaying each round — review the current <DOC>.revN.md into <DOC>.revN.notes.md, commit the closed round, wait for the author's rev(N+1), and repeat until the document is ready for final, a round cap is hit, or only the user can move it forward. Use after rev1 exists, when the user invokes /review-doc-loop <feature> [max-rounds] in a fresh session that did not write the document. Its counterpart is answer-notes-loop in the author's session
Use real run evidence, validation failures, source drift, platform drift, and user feedback to plan and, only after explicit approval, apply structural upgrades to an existing skill. Use when the user asks to improve an existing skill from recent runs, recurring failures, outdated sources, excessive bloat, changed platform behavior, or validated workflow feedback. Do not use to create a brand-new skill or to execute the business workflow itself.
Loads critical project context and recent session history at the start of each session or after `/clear` — reads `.agents/memory/` source-of-truth files, surfaces open next steps from the latest session log, and activates the session-documenter. Triggers on: session start, `/clear`, "load context", "start session", or first message in a new agent session.
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.