Plain SKILL.md files for Claude Code, OpenCode, Codex and SkillMouse. They run the checks, report honestly and stop when they should, with any model.
From the README’s “Install” section. Full options, requirements and troubleshooting are there too.
Open the README$ git clone https://github.com/juw12/skills
# copy any folder from skills/ into:
# Claude Code ~/.claude/skills/
# OpenCode ~/.config/opencode/skills/
# Codex ~/.codex/skills/Improve-until-done with full autonomy: a frozen rubric, a separate judge, keep-best and blind pairwise checks, ending with a stated stop reason.
The original loop for agents that should ask before acting, with an objective self-checker.
Raises any task and pressure-tests a business idea, ending in a verdict computed by a script.
Finds and actually runs a project's build, test and lint commands, and reports pass/fail truthfully.
A short, honest status of running jobs and services that flags stalls.
Audit an app's UI against 8 objective gates and ship a PR, then keep its landing page in sync.
beacon-push, tether-android-qa and push-qa-e2e for real-phone push notification testing.
Subagent versions of harness, heartbeat, delegator, soul and memory-management, plus web-researcher.
123 files in the repo · 12 commits since Nov 2026 · last updated 9 Nov 2026.
Browse the codeGitattributes: force LF so Windows clones keep SKILL.md front matter loadable
Hone: autonomous improve-until-done skill (mandate, run queue, stop reasons); validator checks it
Add ambitious-founder: be-ambitious + founder pipeline in one skill
Agent skills: refine-loop, harness, heartbeat, delegator, soul, memory-management, widgettheme, Push Lab QA
SKILL.md, Any model.
The commands in the “Install” box above come straight from the README. Paste them into a terminal; the README covers options and troubleshooting.
The source is public on GitHub under the MIT licence.
Juwen Lie, an R&D mechanical engineer in Batam, Indonesia, as an independent personal project.
Agent skills that verify instead of claim.