By kim-song-jun
visual-qa scaffold for GitHub Copilot CLI — emits .visual-qa.json + Playwright MCP entry
GitHub Copilot CLI port of /agent-all (intent → plan → wave-dispatch → gate → PR). Uses Copilot's `task` tool for parallel wave dispatch and `store_memory` (scope=repository) for plan persistence. See plugins/harness-floor/skills/agent-all/SKILL.md for the source-of-truth pipeline.
Copilot CLI port of /visual-qa — Playwright MCP capture matrix + per-image LLM analysis + diff vs prior run. Phase 3 uses Copilot's `task` tool for parallel per-page dispatch. See plugins/harness-floor/skills/visual-qa/SKILL.md for the source-of-truth pipeline.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
🇰🇷 한국어: README.ko.md
Agent-first workflows that run themselves. One /agent-init per project; one /agent-all "..." per feature; the agent brainstorms → plans → writes → tests → opens the PR — and keeps iterating until the tests pass — without you babysitting every turn.
Works on Claude Code today, with cross-platform ports for Cursor, GitHub Copilot CLI, VS Code Copilot, Codex CLI, and Gemini CLI. 17 plugins, 5 slash commands, one marketplace.
/agent-init # bootstrap any git repo (Phase A — once per project)
/agent-all "Add Google OAuth" --loop # brainstorm → plan → code → test → PR (Phase C — per feature)
/visual-qa # screenshot every page, LLM design review
/thrift # keep long sessions affordable (auto-summarize, audit)
/explore # codebase map; /explore where Foo → O(1) lookup
/debug "tests flaky 30% of runs" # reproduce → bisect → hypothesize → verify
Three things that make it click:
Project-first scaffolding. /agent-init works on any git repo — Next.js, FastAPI, Rust CLI, monorepo. It detects your stack, picks the right test command, and creates CLAUDE.md + agents + hooks + config in one commit. Same command, every project.
Agent-first execution that preserves your main thread. /agent-all "..." isn't a chat. It runs brainstorm → plan → implement → review → PR as one pipeline, and the implementation/review heavy lifting happens in isolated subagents — their turn-by-turn output never enters your main conversation. A built-in two-layer safety net mandates superpowers:verification-before-completion per implementer + cross-checks at Phase 4 review, so broken code can't sneak into a PR. Your main session stays small (planning + judgment) so the same Claude Code session can keep going for hours without context bloat.
Composable for unattended runs. Three pieces — /agent-all --loop (drives the work), /thrift (compresses what does accumulate in main), /goal (keeps the session alive across iterations) — combine into overnight runs that exit cleanly when CI is green or your cost cap hits. See Self-sustaining workflows.
That's it. The rest of this README is reference material — skim the parts you need.
bin/init.mjs, bin/install.mjs, scripts/install-all.sh, scripts/install-platform.sh)/agent-init, /agent-all, and /explore (HEAD-keyed cache)/agent-all Phase 5 PR creation; without it, /agent-all falls back to --no-pr modeStrongly recommended (the harness composes on top of these — degrades gracefully if missing):
superpowers@claude-plugins-official — foundational skills (brainstorming, writing-plans, subagent-driven-development, verification-before-completion, etc.)context-mode@context-mode — keeps raw tool output out of main conversationSee How this fits with the rest of the Claude ecosystem for details on how they integrate.
First, register the marketplace (once per machine):
/plugin marketplace add https://github.com/kim-song-jun/agent-skill
# Outside Claude Code, in a terminal:
git clone https://github.com/kim-song-jun/agent-skill /tmp/agent-skill
bash /tmp/agent-skill/scripts/install-all.sh
Installs all 5 Claude Code essentials at once via the claude CLI. Run --all for all 17 plugins (CLI-platform siblings too), or --cli=codex|copilot|gemini|cursor for a single platform set.
/plugin install harness-builder@agent-skill
/plugin install harness-floor@agent-skill
/plugin install harness-thrift@agent-skill
/plugin install harness-explore@agent-skill
/plugin install harness-debug@agent-skill
/reload-plugins
(Claude Code's /plugin install accepts only one plugin at a time, so the script in Option A is faster.)
cd my-project
/agent-init
You're done. Try /agent-all "small feature" to see it work.
/plugin update --marketplace agent-skill
npx claudepluginhub kim-song-jun/agent-skill --plugin harness-floor-copilotvisual-qa scaffold for Cursor — emits .visual-qa.json + Playwright MCP entry for .cursor/mcp.json
Run agent-init for Cursor — emits .cursor/rules + .cursor/agents/
Theme B for Cursor — advisory-only via .cursor/rules/thrift.mdc; no programmatic hooks, no cache prime, no metrics
Theme B for Copilot CLI — YAML/JSON hooks under .github/hooks/, store_memory summariser scratch, OpenAI-flavoured cost estimator, Phase 4 cache prime disabled by default (Copilot intermediates the model layer)
Run agent-init for Gemini CLI (a.k.a. 'antigravity') — emits GEMINI.md + .gemini/skills/
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Tools to maintain and improve CLAUDE.md files - audit quality, capture session learnings, and keep project memory current.
Develop, test, build, and deploy Godot 4.x games with Claude Code. Includes GdUnit4 testing, web/desktop exports, CI/CD pipelines, and deployment to Vercel/GitHub Pages/itch.io.
A growing collection of Claude-compatible academic workflow bundles. Covers scientific figures, manuscript writing and polishing, reviewer assessment, citation retrieval, data availability, paper reading, literature search, response letters, paper-to-PPTX conversion, and evidence-grounded Chinese invention patent drafting. Rules are organized as reusable skill folders with explicit workflows and quality checks.
Create new skills, improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, update or optimize an existing skill, run evals to test a skill, or benchmark skill performance with variance analysis.
Unity Development Toolkit - Expert agents for scripting/refactoring/optimization, script templates, and Agent Skills for Unity C# development