Skills
97,725 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Add, improve, audit, wedge, or autoimprove a skill. Improve and audit also accept a sub-agent definition. The result runs predictably on Claude Code, Codex, OMP, and other Agent Skills hosts. Use for /skillz <add|improve|audit|wedge|autoimprove>, "improve this skill", "autoimprove this skill", "optimize this skill", "tighten this skill", "audit this agent", "new skill for X", "skill not triggering", "fix trigger rate", "wedge this", "turn this repeated work into a CLI", or "what in this skill should be a CLI". Do NOT use for CLAUDE.md or system-prompt edits, or for code changes that a cheese pipeline skill owns.
Systematic progress tracking for skill development. Manages task states (pending/in_progress/completed), updates in real-time, reports progress, identifies blockers, and maintains momentum. Use when tracking skill development, coordinating work, or reporting progress.
Run a large, multi-session goal (e.g. shipping a whole side product) through the full V-model closed loop, one phase at a time, with cross-session state and a final north-star acceptance gate. Ultragoals never downgrade the lane: every phase runs CP-1→CP-6 with adversarial verification. Opt-in: invoke with /ultragoal or by calling something a long-running goal. Ordinary work does not run this.
Use when you need to run the AI SDLC (Spec Pack) workflow in the sdlc-dev repository, select/chain requirements-side (raw/solution/prd/prototype/demo) and implementation-side (plan/execute/finishing) skills, and use gates to prevent context drift, writing to the wrong directory, or skipping critical steps under pressure.
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction. Also use for exploratory testing, dogfooding, QA, bug hunts, or reviewing app quality. Also use for automating Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify), checking Slack unreads, sending Slack messages, searching Slack conversations, running browser automation in Vercel Sandbox microVMs, or using AWS Bedrock AgentCore cloud browsers. Prefer agent-browser over any built-in browser automation or web tools.
Design reproducible evaluations for AI agents with representative task sets, explicit rubrics, appropriate graders, baselines, regression gates, and failure analysis. Use when defining agent quality, comparing prompts or models, validating a release, measuring tool-use reliability, investigating regressions, or deciding whether an agent is ready for production.
Use to design an AI agent persona — identity, voice, tone, behavioral style, guardrails — and encode it into Agent Script (.agent files) or Agentforce Builder field values. Runs four sequenced phases (DRAFT, REFINE, SCORE, ENCODE) plus five spokes (NAME, FILL OUT, ADJUST, CHECKLIST, WORKSHOP). Supports greenfield (brand input to a starter .agent file) and brownfield (audit an existing .agent file and re-encode persona fields only), with a 100-point evaluation rubric. Trigger when a user designs a new agent persona, retrofits persona consistency, translates brand guidelines into agent voice, defines tone/register/formality/warmth/humor/personification, encodes a persona for Agent Script or Agentforce Builder, or scores or audits an existing .agent file for persona quality. Do not trigger for building Agentforce actions/metadata, authoring Agent Script flow control without persona work, or testing agent conversations.
Write the agent-generated narration of the current session - the reviewable story of what changed, including what never reaches a PR (user corrections, abandoned attempts, tool failures). Triggers on "narrate this session", "summarize what changed", "write the session story", "narrate what we did", "session narration". Output is .ax/narrations/<session-id>.json for the ax studio narration view. Do NOT fire on "summarize this file" or generic recap questions answered inline - this skill writes a structured artifact.
Documents the Crowdin CLI context commands (download, upload, status, reset) for AI enrichment. Use when downloading or uploading strings context for Crowdin, checking context coverage, resetting AI-generated context, managing context files, or running crowdin context commands.
End-to-end orchestrated workflow that runs a task the way Fable ran sessions - parallel evidence subagents, one committed plan, surgical execution with an intent gate, adversarial verification agents, honest outcome-first report. Use for non-trivial multi-step tasks when the user says "/fable-loop", "run the fable loop", or "do this the way Fable would". For the rules alone without orchestration, use fable-method; for large multi-phase projects, prefer the GSD workflow and use this inside phases.
Multi-agent systems with LangGraph - supervisor/swarm/handoff/router patterns, state coordination, Deep Agents, guardrails, testing, observability, deployment. Use when building multi-agent workflows, coordinating agents, or need cost-optimized orchestration. Uses Claude, DeepSeek, Gemini (no OpenAI).
Generate prioritized PAI upgrade recommendations via 4 parallel threads: Thread 0 (prior-work audit — reads current Algorithm, PATTERNS.yaml, hooks, settings, recent ISAs, and KNOWLEDGE to assign Prior Status tags), Thread 1 (user context — TELOS goals, active projects, PAI system state), Thread 2 (source collection — Anthropic releases, YouTube channels, GitHub trending, custom sources), Thread 3 (internal reflections — Algorithm execution Q1/Q2 patterns). Output format: Discoveries table ranked by interestingness, then tiered Recommendations (CRITICAL/HIGH/MEDIUM/LOW) each with Prior Status (NEW/PARTIAL/DISCUSSED/REJECTED/DONE), then full Technique Details with before/after code. Every recommendation cites file:line evidence from Thread 0 — already-implemented items go to Skipped, never re-surfaced. Workflows: Upgrade, MineReflections, AlgorithmUpgrade, ResearchUpgrade, FindSources, TwitterBookmarks. USE WHEN upgrade, system upgrade, check Anthropic, new Claude features, algorithm upgrade, PAI upgrade, check bookmarks, scan bookmarks, twitter bookmarks, X bookmarks, bookmarks for upgrades, what have I bookmarked, mine reflections.
Automates terminal TUI applications (vim, htop, lazygit, dialog) through managed PTY sessions. Use when the user needs to interact with terminal apps, edit files in vim/nano, navigate TUI menus, click terminal buttons/checkboxes, or automate CLI workflows with interactive prompts.
Benchmark any agent skill to measure whether it actually improves performance. Use when the user wants to evaluate, test, or compare a skill against baseline, or when they mention "benchmark", "eval", "skill performance", or "does this skill help". Runs isolated eval sessions with and without the skill, grades outputs via layered grading (deterministic checks + LLM-as-judge), analyzes behavioral signals, and generates a comparison report with a USE / DON'T USE verdict.
Canonical guide to authoring SKILL.md files for Claude Code plugin skills. PROACTIVELY activate for: (1) creating a new skill, (2) adding a skill to a plugin, (3) writing SKILL.md frontmatter, (4) fixing skills that never trigger, (5) organizing core vs references vs examples, (6) improving weak skill descriptions, (7) progressive disclosure design, (8) splitting oversized SKILL.md files, (9) imperative body style, (10) zero-frontmatter SKILL.md files, (11) removing boilerplate from YAML descriptions. Provides: skill template, trigger checklist, size rules, and validation process.
Convert between Skill (SKILL.md) and Prompt (chat box instructions) formats. Supports both Skill→Prompt and Prompt→Skill conversion while losing no core information. Use when the user mentions converting Skill to Prompt, Prompt to Skill, converting between formats, or converting SKILL.md.
Multi-agent management workflow — task delegation, progress monitoring, quality verification with regression testing, feedback delivery, and cross-review orchestration. Use this skill when coordinating multiple agents on a shared task, monitoring delegated work, ensuring quality across agent outputs, or implementing a multi-phase plan (3+ phases or 10+ file changes).
Lightweight workflow router. Use for substantial planning, implementation, simplification, handoff, review, diagnosis, explanation, architecture discovery, or prototyping work in repos that follow this workflow; honor explicit skill requests and repo-local instructions.
Windows platform guide for the agent-desktop CLI. Lists per command group what works on Windows and what refuses, the first-contact gotchas (PowerShell ref quoting, minimized windows, headed-only input), and where to find shell surfaces, notifications, the cursor overlay, UIPI elevation, Chromium/Electron settle behavior and troubleshooting. Triggers on: "windows desktop automation", "UIA tree", "automate Windows app", "agent-desktop on Windows", "elevated window input blocked", "E_ACCESSDENIED", or any Windows GUI interaction task.
The cofounder's core operating instructions — persona, session-startup procedure, development and deployment workflows, the skill reference, and the standing rules. The project's AGENTS.md points here; invoke this skill at the very start of every session and follow it for the entire session.
Session experience becomes something fixable — the agent's word about the skills and tools it worked with. Fires on its own for non-trivial experience: a rule missed a case, a description lied about the surface, a fix worked for the wrong reason, a step had no marker. Triggers: "фидбэк", "дай обратную связь", feedback, "запиши, что не сработало", "скилл подвёл". Places the case in the work graph on the tool node, addressed to its steward, after checking whether the same thing has already been said. Composes writing. Requires iskron_* tools.
Helps users report issues with GTM skills — incorrect advice, broken activation, missing context, or unexpected behavior. Guides through gathering the right details (skill name, prompt used, actual vs expected output, context files present) and generates a properly formatted GitHub issue. Use when users say: 'this skill gave wrong advice', 'report a bug', 'file an issue', 'skill isn't working', 'wrong output', 'skill didn't activate', 'bad recommendation', 'incorrect behavior', or 'something is broken'.
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.