Skills

skills

Agent Skills published by @pjankiewicz. Drop-in references for AI coding agents (Claude Code, Cursor, Copilot CLI, Codex, Cline) following the open Agent Skills format.

Skills in this repo

open-multi-agent

A skill for @jackchen_me/open-multi-agent — the TypeScript-native multi-agent orchestration framework that turns a goal into a task DAG. Teaches the agent which of the three execution modes to pick (runAgent / runTeam / runTasks), which built-in tools to attach, how to wire MCP servers, how to use Zod-validated structured output, how to mix providers in one team, and which production controls actually matter.

npx skills add pjankiewicz/skills@open-multi-agent

To install every skill in this repo at once:

npx skills add pjankiewicz/skills

Browse the directory at skills.sh.

ux-review

A per-screen, PASS/FAIL heuristic rubric (~55 criteria across 9 sections) for auditing the interaction quality of a mobile (iOS/Android) or web UI — pressed/focus/disabled states, empty/loading/error states, touch targets, platform conventions (Apple HIG / Material 3), WCAG 2.2 AA contrast, system-status feedback, and honest copy. Hunts works-but-feels-wrong.

npx skills add pjankiewicz/skills@ux-review

app-bug-sweep

A fan-out methodology for finding the defects that ship — the integration gaps a compiler and unit tests never catch: dead/no-op controls, capability built on the backend but never wired to a screen, inert gestures, silent data-loss footguns, and dead settings / fake data / cross-platform decode-parity gaps. Runs as parallel read-only scans, one per failure class, with a suppression list so each reports only net-new findings. Pairs with ux-review (one hunts broken, the other feels-wrong).

npx skills add pjankiewicz/skills@app-bug-sweep

one-item-per-file

A code-as-data authoring methodology for any non-trivial module: write it in three phases — (1) define the data structures, (2) define the function signatures so the skeleton compiles, (3) implement the functions — with strict one item per file (every type and every free function in its own file; a module is a directory of them). Isolated items parallelize cleanly and related ones batch; compile gates between phases catch a mistake where it was made. Makes code precise to write, parallelize, and maintain.

npx skills add pjankiewicz/skills@one-item-per-file

faithful-port

Porting/translating code from a cited reference (C++ → Rust, GLSL → WGSL, one engine's algorithm into yours) where faithfulness is the goal. Builds on one-item-per-file and adds reference fidelity: transcribe each reference function 1:1 into its own file (never paraphrase or consolidate — that silently drops branches and edge-cases), resolve needed types/functions against the existing codebase (reuse vs declare-missing), batch implementations by how functions are collocated in the reference (the speed/cost↔coherence dial), and verify each item against the reference with an independent oracle — not a self-check, and not an aggregate metric that hides a localized defect.

npx skills add pjankiewicz/skills@faithful-port

What the open-multi-agent skill covers

• Three run modes — runAgent (single), runTeam (goal-driven coordinator), runTasks (explicit DAG).
• Tool wiring — built-in tool presets, custom tools via defineTool + Zod, opt-in delegate_to_agent.
• MCP — connectMCPTools from the @jackchen_me/open-multi-agent/mcp subpath.
• Providers — Anthropic, OpenAI, Azure, Gemini, Grok, DeepSeek, MiniMax, Qiniu, Copilot natively; Ollama / vLLM / LM Studio / OpenRouter / Groq via OpenAI-compatible baseURL.
• Structured output — outputSchema (Zod) on AgentConfig → result.structured.
• Observability — onProgress events, onTrace spans, renderTeamRunDashboard static HTML.
• Production controls — maxTurns, maxTokenBudget, timeoutMs, contextStrategy, loopDetection, compressToolResults, task retries, delegation depth caps, onApproval gate.
• Common pitfalls — ESM-only, Ollama placeholder API key, local-server tool-call streaming quirks, peer-dep loading.

Skills are read on demand by the agent — they do not run code, install dependencies, or modify your project.

Tools shipped alongside

agent-farm/

A small, generic live dashboard for multi-agent runs. Watches a runs directory and renders any agent activity that follows the [agent-farm event format](./agent-farm/EVENTS.md). Intentionally domain-agnostic — anything that emits well-formed JSONL with a taskId shows up. Pairs well with the open-multi-agent skill but doesn't depend on it.

See [agent-farm/README.md](./agent-farm/README.md) for usage. Install / run from the repo:

cd agent-farm && npm install && npm run dev
# dashboard at http://localhost:5180/, watching ./agent-runs/ by default

Layout

skills/
  /
    SKILL.md           # the skill itself
agent-farm/             # optional sidekick dashboard for multi-agent runs

Standard npx skills / skills.sh skill layout — one directory per skill under skills/, each with a SKILL.md carrying YAML frontmatter (name, description).

Compatibility

• Claude Code (~/.claude/skills/)
• Cursor, VS Code, GitHub Copilot, Cline, Codex, and any other agent that consumes the open Agent Skills format.

Source of truth

The open-multi-agent skill summarises the public API of @jackchen_me/open-multi-agent as of the version pinned at publish. For exact current signatures and edge cases, the upstream sources of truth are:

• Repo:
• npm: @jackchen_me/open-multi-agent
• CLI docs: docs/cli.md
• Shared memory: docs/shared-memory.md
• Context strategies: docs/context-management.md

If something here drifts from the upstream API, file an issue or open a PR.

License

[MIT](./LICENSE). Skills here describe third-party packages; those packages remain the property of their authors.

Not affiliated

These are independent community skills. They are not maintained by, endorsed by, or otherwise affiliated with the upstream projects they document.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: pjankiewicz
• Source: pjankiewicz/skills
• License: MIT

Install and usage instructions live in the source repository linked above.