JS

jmkelly/skills

效率工作流

本页由 AI 翻译,技术细节请以原文为准。 查看原文

0 stars 质量 53 趋势 20MIT

pi agent skills: quality-loop, writing-for-agents

概览

Agent skills for the pi LLM harness, as installed at ~/.pi/agent/skills. Per-component MIT — see LICENSE (repo, © James Kelly), quality-loop/LICENSE, and writing-for-agents/LICENSE.

README

以下为原文 README 节选,完整内容请查看原文。

skills

Agent skills for the pi LLM harness and other Agent Skills clients, installed at ~/.agents/skills or your harness’s skills folder.

kelso-mode is the entry to the set. Trigger it with kelso or /kelso-mode. It routes each request to the skill that owns it, names the principles that shaped the result, and runs the matched playbook. Every other skill in the repo is bundled with it.

Skills

Skill What it does When to reach for it License
kelso-mode Trigger routing, the principles catalog, subagent defaults, autonomy, reply and comment style, and beads-based playbook mechanics. Asked to work in kelso mode. MIT © James Kelly
quality-loop Deterministic quality gates for .NET and Python repos — CRAP < 10 per method/function, metrics (radon / Dependably.CodeMetrics), Stryker mutation testing — plus a two-agent loop (verifier + implementor) that drives the gates green. Stack auto-detected; gates exit 0/1, offenders land in crap-queue.md / metrics-queue.md / stryker-queue.md. Asked to run the quality loop or an audit; a queue reports offenders; the gates or their tooling are mentioned. MIT © James Kelly
writing-for-agents Reference for writing documents agents consume: skills, AGENTS.md / CLAUDE.md, docs reached by a pointer. Covers context pointers, the two loads (context vs cognitive), and information hierarchy; see SKILL-MECHANICS.md for skill packaging. Upstream: mattpocock/skills. Creating or editing skills, or modifying AGENTS.md / CLAUDE.md. MIT © Matt Pocock
jev-decisions Decision-only layer for OpenCode and Paseo: bound each decision, call jev_decide for a typed answer, apply it, and write the artifact. Ships the MCP server, harness plugins, and an LLM-judge eval. An agent is about to choose, rank, approve, reject, include, exclude, continue, or stop; implementation, scope, review, planning, advice, or next-step work. MIT © James Kelly

Bundled leaves

kelso-mode bundles 29 leaf skills, each also invocable on its own:

  • arena runs N parallel candidates at one task, picks a base, and grafts the strongest parts of the losers into it.
  • show-me-your-work keeps a reviewable decision trail for long or unattended runs.
  • benchmark-checklist vets a measured performance number before you report or act on it.
  • automate-me drafts or refreshes a personal -mode skill from the user’s working conventions and recent transcripts.
在 GitHub 查看完整 README

推荐工具

换一个关键词,或者移除筛选条件。

安装

npx skillfish add jmkelly/skills
新手?先看 mcpskill.org 教程 →