universal-skills

Scaffold a complete agent-package project with all config files, Docker infrastructure, MCP server, A2A agent, and API client stubs. Use when creating a brand-new agent-package from scratch, bootstrapping a new MCP/agent/api-client project, or when the user says "create a new agent package". This delegates domain-specific implementation to existing skills (api-client-builder, mcp- builder, agent-builder, skill-graph-builder). Do NOT use for modifying an existing agent package — use the individual skills directly.

Browser automation CLI for AI agents using the agent-browser tool. 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.

Guide for building scalable Pydantic AI agents. Use this skill when the user wants to create a new agent package or modify an existing agent's architecture, to ensure it follows the standardized agent patterns using `agent-utilities`.

The `agent-spawner` skill provides the ability to dynamically create a new Pydantic AI agent configured with an external MCP server toolkit. This is extremely useful when an agent wants to spin up a sub-agent with specialized tools defined in an `mcp_config.json`, and interact with that agent immediately without modifying the parent agent's state or toolset.

Consolidated skill for managing, dispatching, and orchestrating other agents via the agent-manager CLI, as well as workflows for A2A communication, orchestration, and parallel subagent dispatch.

Use when executing implementation plans with independent tasks in the current session

Generate or update a robust AGENTS.md file (the "README for AI coding agents") for a project. This skill ensures your repository is optimized for AI tools (like Copilot, Cursor, and Windsurf) by providing structured context, commands, architecture diagrams, and coding standards. It automatically detects project metadata from pyproject.toml and generates a project tree view.

Guide for creating API clients using the established patterns in the agent- packages repository (like servicenow-api and gitlab-api). Use this when tasked with building a new API integration or creating client classes for REST or GraphQL APIs from OpenAPI specs or documentation.

Runs one propose-only self-evolution "golden loop" cycle over the Knowledge Graph. Use when the user wants to "run the research loop", "evolve the knowledge graph", "find and address open topics", "propose specs/teams from the KG", or "self-improve the codebase". Intake of unresolved topics → acquire related sources → ADDRESSES-resolve (converge) → optional distil spec DRAFTS into .specify/ → synthesize a team PROPOSAL. Never auto-merges or executes anything — every artifact is a draft/proposal a human reviews.

Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services using the approved FastMCP Python standardized template.

Universal MCP client for connecting to any MCP server directly from a skill. This replaces the old multi-agent pattern where the parent agent held all MCP tools — instead, each skill spawns its own MCP connection with only the tools it needs.

Install a pre-configured MCP config file into various agent tool directories. Use when the user wants to install an mcp_config.json into Windsurf, Claude Code, OpenCode, Antigravity, Devin, Codex, or a custom agent path. It handles merging with existing configurations.

Author and validate canonical system-prompt JSON blueprints for agent-utilities agents and agent-packages. Use when creating or fixing a main_agent.json / a prompts/*.json system prompt, when an agent-package needs its system prompt, or when a prompt must conform to the StructuredPrompt canonical schema. Produces schema-valid blueprints with the body in instructions.core_directive plus schema_version/source/skills/tools, and validates them against the one shared validator (validate_canonical). Do NOT use for content-generation templates or for authoring skills (use skill-builder).

Transform any documentation source — a website, a PDF/Office file, a local directory, a single URL, a REST/DB/MCP connector, freshly generated text, or a Knowledge-Graph subgraph — into a polished, standardized agent skill-graph (SKILL.md + reference/ manual + machine-readable index.json + optional LLM-distilled OVERVIEW.md) with a sources.json provenance/freshness manifest, hybrid Knowledge-Graph + ontology enrichment, and keep-updated refresh/restyle/rebuild loops. The recipe to capture an entire reference manual as a reusable skill and create more.

Tool and instructions to interactively brainstorm, structure, verify, and scaffold structured workflow-based skills. Use when the user or agent needs to create a new workflow skill under `universal_skills/workflows/`, define topological step lists, or analyze dependencies of a complex agent process. Do NOT use for standard non-workflow skills or simple file updates.

Run Claude Code unattended ("while you sleep") to drive the graph-os Loop engine — feature extraction + innovation distillation — behind a governance-derived permission fence. Use when the user wants to "run the loop overnight", "let Claude work unattended", "run while I sleep", "set up an autonomous build session", or "drive the evolution loop without babysitting". Verifies the fence is active (settings.json allow/ask/deny + .claudeignore, no bypassPermissions), then loops graph_loops(action="run") committing per productive cycle, halts on any ask/deny verdict instead of auto-approving, and writes a morning summary to MEMORY.md for review.

Comprehensive code analysis and enhancement skill that performs 31-domain deep- dive reviews of any codebase. Covers project analysis, dependency audit, changelog audit, codebase optimization, security analysis, test coverage, pytest quality grading, test execution, pre-commit compliance, documentation governance, directory organization, language detection, UI/UX heuristics, architecture review, concept traceability, linting, vulnerability scanning, environment variable scanning, runtime and scale profiling (memory, startup, instances-per-GB), brainstorming, multi-project orchestration, cross-project integration, and actionable reporting with standardized 0-100 grading and SDD handoff integration. Language-agnostic: supports Python, Go, Node, Rust, Java. Use when tasked with "auditing", "optimizing", "updating", "improving", "reviewing", "grading", or "enhancing" an agent, repository, or codebase. Replaces self-improver.

Ultra-compressed communication mode. Cuts token usage ~75% by dropping filler, articles, and pleasantries while keeping full technical accuracy. Use when user says "efficient mode", "compressed mode", "less tokens", "be brief", or invokes /efficient.

Create and restore comprehensive context handoff documents for agent session transfers. Use when context is getting full, a major task milestone is complete, work is pausing, or when resuming from a previous session. Triggers include "create handoff", "save state", "context is full", "I need to pause", "load handoff", "resume where we left off", or proactively after 5+ file edits or complex multi-step work. Do NOT use for project documentation — this is for agent-to-agent context preservation.

Guide for creating effective skills for the universal-skills package. Use this when users want to create a new skill or update an existing skill according to the project's standards.

Install and wire the whole fleet contribution surface into an agent tool in one step: skills, skill-graphs, skill-workflows, ontologies, and system prompts — plus generating/patching that tool's mcp_config.json to wire graph-os (local stdio via uvx, or a remote instance) and every auto-detected agents/* MCP server. Supports installing from pip-installed providers or directly from an agents/* project checkout. Use when the user wants to install skills or ontologies or prompts into Windsurf, Claude Code, OpenCode, Antigravity, Codex, Devin, Cursor, or a custom agent path, or wants an agent tool wired up end-to-end (skills + MCP servers) in one command. Do NOT use for creating or editing skill/ontology/prompt content — use skill-builder / skill-workflow-builder / skill-graph-builder for that.

Creates generative algorithmic art using p5.js with seeded randomness and interactive HTML artifacts. Use when the user requests generative art, algorithmic art, computational aesthetics, flow fields, particle systems, or creative coding. Output is self-contained interactive HTML. Do NOT use for UI design, data visualization charts, or photo editing.

Creates comprehensive brand identity systems including logo concepts, color palettes, typography scales, imagery direction, and voice/tone guidelines. Use when establishing a new brand, evolving an existing brand identity, or creating brand standards documentation. Do NOT use for general graphic design without a brand strategy context.

Creates visual designs using HTML Canvas or SVG APIs for banners, social media assets, posters, or UI components. Use when the user requests programmatic graphic design, visual composition, or custom image generation without using a photo editor. Do NOT use for photo realistic image generation or general web UI layouts unless it specifically requires canvas/SVG.

Creates and manipulates creative media assets including animated GIFs, social media videos, image format conversions, and basic video editing workflows. Use when the user requests an animated GIF from images, video clipping, format conversion (PNG to WebP), or creative media processing. Do NOT use for generic data processing or non-creative file management.

Universal automated testing atomic skill. Discovers environment, executes test suite with timeouts, and compiles execution outcomes.

Model AND decide software architecture. (1) Generate C4 documentation with Mermaid (Context/Container/Component/Deployment) — triggers "architecture diagram", "C4 diagram", "system context", "container/component/deployment diagram", "document architecture". (2) Reason about architecture DECISIONS grounded in canonical references (Hohpe, Richards/Ford, Hard Parts, Tar Pit, Kleppmann DDIA, DDD, Team Topologies) — triggers "should we split/merge/extract this service", "monolith vs microservices", "service/bounded-context boundaries", "saga / distributed transaction / consistency model", "draft an ADR", "fitness functions". Do NOT use for general flowcharts or ERD (use mermaid-diagrams), code-level refactoring, or framework picks (React vs Vue).

Generates code walkthroughs

Creates or updates constitution.md with project governance rules and principles.

Disciplined diagnosis loop for hard bugs and performance regressions. Reproduce → minimise → hypothesise → instrument → fix → regression-test. Use when user says "diagnose this" / "debug this", reports a bug, says something is broken/throwing/failing, or describes a performance regression.

Interactive codebase explanations

Enhanced manual testing workflows

Executes tasks and tracks progress natively with Tasks Pydantic models.

Generates spec.md with user stories and acceptance criteria. Replaces product- management.

Interactive Spec Intake Wizard atomic skill. Prompts the user with structured fields, clarifies ambiguity, validates inputs, and outputs parsed JSON parameters.

Polished verifier skill with drift report and CHECKLIST.md output for spec-kit parity

Generates executable tasks.md with dependency graph and [P] parallelism markers

Strict Test-Driven Development (TDD) methodology guide using the Red-Green- Refactor cycle. Use when implementing features or fixing bugs with TDD, or when the user mentions TDD, test-first development, or comprehensive test coverage requirements. Do NOT use for pure design discussions, documentation tasks, or research-only work.

Parallel execution workflow for code enhancement and multi-domain deep-dive auditing using the Unified Parallel Engine

Unified ecosystem standardization workflow for the agent-packages repository. Audits all projects for structural drift, generates missing standardized files, migrates environment variables, extracts MCP subdirectories, and creates standard test suites. Produces a comprehensive drift report with per-project scorecards. Use when onboarding a new agent, auditing ecosystem compliance, running "standardize the ecosystem", "check for drift", "onboard new agent", or "ecosystem audit".