skills
Built skills: Agent skills I wrote for Claude Code, OpenCode, Codex and SkillMouse
Agent skills I wrote for Claude Code, OpenCode, Codex and SkillMouse: agents that verify instead of claiming (refine-loop, harness, heartbeat, Push Lab QA...)
- Repo: https://github.com/juw12/skills (public)
- Language: Python
- Topics: agent-skills, ai-agents, claude-code, opencode, skills
- Started: 2026-11-04 · last push: 2026-11-06
From the README
Agent skills I wrote for Claude Code, OpenCode, Codex and SkillMouse. Plain SKILL.md files that work with any model.
The common thread: agents that verify instead of claiming. They run the checks, report honestly and stop when they should.
Changes
- 2026-11-20:
/honeholds every project to four versions: a Windows and a Mac download and an Android APK in a GitHub release, and for iPhone the live web app on the home screen.platforms.pyis the gate and the README has to name all four (d8552f3). First reading of the profile: 4 of 92 projects have all four (Glow, Haru, Mint, Vitals); Windows 20, Mac 9, iPhone 46, Android 14. It checks that a version is published, not that it runs. - 2026-11-20:
/honeships a finished project to every place it shows up: GitHub, the profile, the portfolio, Obsidian and Cairn, each as a run-queue item with its own gate (ship.md). Before a pushpush_as.pyreads which account Cairn is set to push as and never switches it; repos stay underjuw12unless the request names another owner.featured.pyscores the profile's projects from facts (test files, releases and downloads, a live demo that answers, commits, stars, CI, Cairn's ranking note), orders the Featured cards and names swaps and pins (a94ff8f). Not yet used in a real/honerun; creating a repo and inviting the pushing account are checked against a simulated GitHub only.
See also: GitHub repos