Skills
96,981 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Use this skill when users want to execute this workflow when a user wants to test how a different LLM model would handle existing Agentforce agent conversations. Trigger when users mention agentforce, offline, experimentation
Use this skill for any explicit wiki mutation in ai-memory: saving durable or time-bounded project knowledge, recording a rule or annotation, updating a note, or deleting a memory page. Trigger by semantic intent rather than exact wording; routine session capture is not a durable-page request.
Use this skill for any ai-memory knowledge-base maintenance request: consolidating observations, reviewing session lessons, proposing durable learnings, auditing or linting the wiki, finding contradictions, pruning stale memory, or running auto-improvement. Trigger by semantic intent rather than exact wording.
Use this skill for any request whose goal is read-only retrieval from ai-memory: project history, prior context, decisions, rules, gotchas, recent activity, full wiki pages, or status/briefing. Trigger by semantic intent rather than exact wording, including when ai-memory is not named.
Use this skill for any request to install, refresh, repair, inspect, or remove ai-memory's agent-facing routing: managed instruction snippets, Agent Skills, CLAUDE.md/AGENTS.md integration, or local/global skill roots. Trigger by semantic intent rather than exact wording.
Vercel AI SDK v7 development and migration. Use when building or upgrading AI SDK 7 apps, especially ToolLoopAgent, WorkflowAgent, HarnessAgent, Claude Code/Codex/Pi harnesses, runtimeContext, toolsContext, toolApproval, telemetry, reasoning, file or skill uploads, realtime, video generation, or v6-to-v7 breaking changes. For AI SDK v6 code use ai-sdk-6; for version discovery and general doc lookup use ai-sdk.
Compiles any research input — PDF papers, GitHub repositories, experiment logs, code directories, or raw notes — into a complete Agent-Native Research Artifact (ARA) with cognitive layer (claims, concepts, heuristics), physical layer (configs, code stubs), exploration graph, and grounded evidence. Use when ingesting a paper or codebase into a structured, machine-executable knowledge package, building an ARA from scratch, or converting research outputs into a falsifiable, agent-traversable form.
Autonomously improve a real artifact (code, training recipe, agent harness, data pipeline, prompt) against an objective and an evaluator, using Hypothesis Tree Refinement (HTR) from the Arbor paper. Use this whenever someone wants to iteratively optimize something over many experiments without overfitting — e.g. "get my model's eval score up", "improve this agent/harness", "tune this pipeline", "beat the baseline on this benchmark", "run a search over approaches and keep the best", "do an MLE-bench / Kaggle-style optimization", or any long-horizon "make this artifact better and don't just memorize the dev set" task. Trigger it even when the user doesn't say "Arbor" or "hypothesis tree" but describes repeated experiment-and-evaluate loops, branching exploration of competing ideas, or worries about a dev/test gap. Runs Claude itself as the coordinator with subagent executors in isolated git worktrees; for the standalone `arbor` CLI tool see references/arbor-upstream.md.
Core ArchDev workflow — use for anything involving ArchDev. Covers CLI setup, upgrades, login, and model access; archdev.json validation (check); repo onboarding and readiness (repo status); activity taxonomy mapping (repo map); harness hooks (repo hook setup); build-event reporting (log post, log post --event); prior lessons (log messages, log search, capability-gated log knowledge); artifact findings (log finding); lifecycle posts — start, lesson, abandoned, done, handoff, question (log --kind); agent activity (repo monitor); live Ask dashboards (archdev ask); public file sharing (archdev share); mandatory PR CI and risk follow-through; published PR risk/scope (inspect metadata); and repository rules (rules, rules add, rules edit, rules disable). Load at session start whenever the archdev CLI is installed, the repo contains archdev.json, or the task touches plans, tasks, sessions, commits, PRs, or harness hooks.
Drive Archon through its CLI: run AI workflows on a repo, watch and answer runs that need a decision, manage the SDLC pack's queues (intake, build queue, merge queue, discoveries), set up Archon or change its config, and author and validate workflows. Use when the user says "use archon", "run archon", "archon workflow", "fix issue #N with archon", "have archon review this PR", "what's running / check run <id>", "approve/reject/answer/cancel/resume that run", "run the build queue", "drain discoveries", "set up archon", "configure archon", "change my archon config", "create a workflow", "author a workflow", or asks how to write a better Archon prompt. NOT for: doing the coding work yourself. Archon delegates it to isolated runs.
Write a new SecSkills skill end to end—choose the plugin bucket and skill tier, write a description that triggers correctly without stealing traffic from siblings, add the required sections, register the skill in ttp-index.json, add routing eval cases including negative traps, and run the three validators. Use when adding a skill to this repo, splitting or merging existing skills, fixing a skill that triggers on the wrong requests, or when a new skill fails validate.py, sync_attack.py, or run_evals.py.
Auto-review pipeline. Chains the four plan-* reviews (plan-deep-review, plan-design-review, plan-eng-review, plan-devex-review) at full depth with intermediate AskUserQuestion calls auto-decided by 6 principles. Surfaces taste decisions and user challenges at a single Final Approval Gate. Sequential Deep → Design → Eng → DX. Each phase optionally gets an independent Claude subagent voice for cross-model consensus. Use when asked to "autoplan", "run all the plan reviews", "auto-review this plan", or when the user wants comprehensive plan vetting without sitting through every intermediate question.
Observe the user's screen via screenpipe, detect repeated research workflows, match them against existing research-agent-skills, and draft new skills (or composition recipes that chain existing ones) for the patterns not yet covered. Use when the user asks to analyze their recent work and propose skills based on what they actually do. Requires the screenpipe daemon (https://github.com/screenpipe/screenpipe) running locally on port 3030 — the skill has no other data source and will refuse to run if screenpipe is unreachable. All detection runs locally; only redacted cluster summaries reach the LLM.
This skill helps an LLM pick the right AxAgent context tool for a job - contextMap for recurring corpora, contextPolicy presets for within-run trajectory compaction, agent.optimize for offline GEPA instruction/demo tuning, agent.playbook for an evolving context playbook (offline evolve + online update), and recall/memories + skills for per-turn retrieval. Use when the user asks "which context feature should I use", confuses contextMap with contextPolicy or memory, or wants a decision guide for long-context agents. For contextPolicy/contextMap codegen use ax-agent-rlm; for recall/skills use ax-agent-memory-skills; for agent.optimize or agent.playbook use ax-agent-optimize.
This skill helps an LLM generate correct playbook code using @ax-llm/ax. Use when the user asks about playbook(), AxPlaybook, context playbooks, evolving context, ACE / Agentic Context Engineering, agent.playbook(), or growing/applying task knowledge offline and online with evolve() and update().
Upgrade blueprint structure to the latest format version. Use when the user asks to "upgrade blueprint", migrates between format versions (v1.x→v2, v2→v3, v3.0→v3.1/3.2/3.3), adds the task registry introduced in v3.2, enables monorepo workspaces in v3.3, or runs batch upgrades across repos with --non-interactive / -y.
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
Automatically back up OpenClaw core configuration, skill files, and memory data to a mounted TOS drive, stored by date, automatically skipping dependency directories and executable files. Use this skill when you need to create or restore an OpenClaw environment backup.
Compress natural language memory files (CLAUDE.md, todos, preferences) into caveman format to save input tokens. Preserves all technical substance, code, URLs, and structure. Compressed version overwrites the original file. Human-readable backup saved as FILE.original.md. Trigger: /caveman-compress FILEPATH or "compress memory file"
Inject structured reasoning blocks into prompts to encourage step-by-step thinking. Use when tasks require complex problem-solving, multi-step reasoning, planning, analysis, or when you need the model to show its work and reflect on decisions before providing answers.
PERSONAL APP ARCHITECT - Strategic development orchestrator for personal productivity applications. Analyzes project context, makes architectural decisions for single-developer projects, delegates to specialized skills, and ensures alignment between user experience goals and technical implementation. Optimized for personal apps targeting 10-100 users.
Comprehensive knowledge for using the Bash tool in Claude Code effectively. This skill should be used when orchestrating CLI tools, configuring hooks, setting up automation workflows, managing git operations, handling multi-command patterns, or encountering Bash tool errors. Covers: PreToolUse hooks, command chaining patterns, git workflow automation, CLI tool integration, custom commands (.claude/commands/), security configurations, allowlisting, session persistence, output handling, error prevention, and troubleshooting common issues. Use when: setting up Claude Code hooks, configuring bash permissions, creating custom commands, automating git workflows, orchestrating multiple CLI tools, debugging bash command failures, implementing security guards, logging command execution, or preventing dangerous operations.
Pull the image sitting on the user's clipboard into the conversation so you can actually look at it. A local bridge exposes the clipboard as a per-user Unix socket ($HOME/.clipboard.sock; CLIPBOARD_SOCKET overrides); this skill retrieves the bytes, saves them as a real image file, and reads that file into context. Use it whenever the user points at something visual you cannot see — "look at my clipboard", "I just took a screenshot", "check out this error", "why does this look wrong", "here's the design", "see the paste" — including when they only imply they copied something, and including when they describe on-screen content without giving you a file path. Reach for it instead of guessing, asking them to describe the picture, or asking them to save the file somewhere first.
Implement Corrective RAG (CRAG) and Self-RAG for reliable, self-healing retrieval systems. Use this skill when building reliable RAG, preventing hallucinations, implementing retrieval evaluation, or adding self-correction to RAG pipelines. Activate when: corrective RAG, CRAG, self-RAG, hallucination prevention, retrieval evaluation, RAG reliability, self-healing RAG, document grading.
Automate Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify, etc.) using agent-browser via Chrome DevTools Protocol. Use when the user needs to interact with an Electron app, automate a desktop app, connect to a running app, control a native app, or test an Electron application. Triggers include "automate Slack app", "control VS Code", "interact with Discord app", "test this Electron app", "connect to desktop app", or any task requiring automation of a native Electron application.
Capture how the system feels — energy, momentum, burnout, breakthrough. Emotional intelligence for Oracle-human collaboration. Use when user says 'feel', 'how are we', 'energy check', 'burnout', 'momentum', or wants emotional awareness of the work.
Execute prompts as the GLM seat through the Pi CLI on OpenRouter, with Z.AI's GLM 5.3 Flash pinned per invocation (z-ai/glm-5.3-flash). Use when users explicitly request GLM execution, when a workflow needs a GLM-labelled seat, or when a cross-runner workflow selects GLM as a complementary provider.
Write the current session's work into a versioned, self-contained handoff document in .agents/handoffs/ and point to it from AGENTS.md, so another session, agent, or person can resume it. Use when the user asks for a handoff, or to write up where the work stands for another session, agent, or person.
Helixa — Onchain identity, reputation, and Cred Scores for AI agents on Base. Use when an agent wants to mint an identity NFT, check its Cred Score, verify social accounts, update traits/narrative, query agent reputation data, check staking info, or search the agent directory. Supports SIWA (Sign-In With Agent) auth and x402 micropayments. Also use when asked about Helixa, AgentDNA, ERC-8004, Cred Scores, $CRED token, or agent identity.
Use for all repository edits and their planning, including simple code, documentation, configuration, and test changes. Also use to choose the next task from a tracker, continue work, recover context, or split implementation into tasks. Review-only and publication-only requests are outside this workflow.
Run the guided, end-to-end loop that makes one Duvo Agent better. Use when the user wants to "improve this Agent", "make my Agent better", "help me tune this Agent", "this Agent isn't good enough", "fix this Agent based on its Runs", or asks you to take an Agent from where it is to a concrete, applied improvement — not just a read-only audit. You drive the conversation: pin the Agent, survey its whole setup (AOP, Connections, Skills, triggers, memory, recent Runs), agree on what to base the improvement on (the user's feedback or the evidence in the Runs), propose a Run range and confirm it, get the grounded findings from the workflow-debugger / run-debugger skills, propose concrete changes, and — on the user's yes — apply them (AOP via aop-writer as a new Build, config with your tools). The boundary against workflow-debugger: that skill is the read-only audit; this skill is the interactive loop that ends in an applied change and calls workflow-debugger to do the analysis.
Run the guided, end-to-end loop that makes one Duvo Queue — and the producer→consumer workflow around it — better. Use when the user wants to "improve this Queue", "this Queue keeps backing up, fix it", or take a Queue to a concrete, applied improvement — not just a read-only audit. You drive the conversation: pin the Queue, survey its whole shape (producers and consumers, backlog and case state, both Agents' AOPs at the seam, triggers and concurrency, the Runs flowing through), agree what to base the improvement on (the user's feedback or the evidence in the Runs/backlog), propose a Run/case range and confirm it, get grounded findings from the workflow-debugger / run-debugger skills, propose concrete changes, and — on the user's yes — apply them (AOPs via aop-writer as new Builds, topology and triggers with your tools). The boundary vs workflow-debugger: that skill is the read-only audit; this is the interactive loop that ends in an applied change. For a single standalone Agent, use improve-agent instead.
Reviewer-gated iterative fleet for headless `claude -p`, `codex exec`, or `pi -p` workers that run in cycles until a designated reviewer approves the output. Use when the work needs multiple rounds of iteration with a quality gate — a reviewer worker reads all worker logs, writes a verdict (lgtm | iterate | escalate), and the orchestrator decides whether to continue, pause, or stop. NEVER kills or restarts workers automatically; the operator owns all kill/pause decisions.
Use Judgment for agent tracing, evaluations, Code judges, datasets, and monitoring. Use when integrating Judgment or judgeval, adding tracing to agents/workflows, creating evaluations or scorers, debugging traces, or looking up Judgment SDK usage and docs.
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.