{
  "markdown": "<div align=\"center\">\n\n# HarnessKit\n\n**Adaptive harness for vibe coders — detect, configure, observe, improve**\n\n[![Version](https://img.shields.io/badge/version-0.4.3-blue)]()\n[![Tests](https://img.shields.io/badge/tests-122%20passing-green)]()\n[![License](https://img.shields.io/badge/license-MIT-yellow)]()\n\n[한국어](README.ko.md) | English\n\n</div>\n\n---\n\n## Overview\n\nHarnessKit is a [Claude Code](https://docs.anthropic.com/en/docs/claude-code) plugin that wraps your development workflow in an adaptive harness. It detects your stack, applies guardrails matched to your skill level, observes your sessions, and continuously improves itself — all without wasting context tokens.\n\n```\n┌─────────────────────────────────────────────────┐\n│                  HarnessKit                      │\n│                                                  │\n│   Detect ──▶ Configure ──▶ Observe ──▶ Improve   │\n│     │           │             │           │      │\n│  language    presets      session       insights  │\n│  framework   guardrails   hooks        proposals  │\n│  tooling     briefing     metrics      auto-apply │\n│                                                  │\n│          Zero-Token Shell Hooks                  │\n│          ── bash + jq, no LLM cost ──            │\n└─────────────────────────────────────────────────┘\n```\n\n## Features\n\n| Stage | What it does | How |\n|-------|-------------|-----|\n| **Detect** | Auto-detects repo language, framework, test framework, linter, package manager | Zero-token shell script — no LLM calls |\n| **Configure** | Applies one of 3 presets (beginner / intermediate / advanced) | Controls guardrails depth, briefing detail, dev hooks |\n| **Observe** | Tracks errors, tool usage, plugin effectiveness per session | Shell-based hooks (session-start, guardrails, post-edit-lint, post-edit-typecheck, session-end) |\n| **Improve** | Analyzes session data, proposes skill/agent/hook/rule improvements | `/harnesskit:insights` analyzes, `/harnesskit:apply` executes |\n\n## Install\n\nIn Claude Code, open the plugin menu and add the marketplace, then install:\n\n```\n/plugin  →  Marketplaces  →  Add  →  ice-ice-bear/harnesskit\n/plugin  →  Discover  →  harnesskit  →  Install for you (user scope)\n```\n\nAfter installation, run `/reload-plugins` to activate. To update after new releases:\n\n```\n/plugin  →  Installed  →  harnesskit  →  Update now\n```\n\n## Quick Start\n\n```bash\n# 1. Initialize harness for your repo\n/harnesskit:setup\n\n# 2. Code normally — hooks observe your session automatically\n\n# 3. After a few sessions, review insights\n/harnesskit:insights\n\n# 4. Apply recommended improvements\n/harnesskit:apply\n```\n\nThat's it. HarnessKit adapts to your workflow over time.\n\n## Commands\n\nHarnessKit provides 11 skills, accessible as slash commands:\n\n| Command | Description |\n|---------|-------------|\n| `/harnesskit:setup` | Initialize harness — detect stack, choose preset, generate config |\n| `/harnesskit:insights` | Analyze session data, propose improvements |\n| `/harnesskit:apply` | Review and apply proposals (with optional A/B eval) |\n| `/harnesskit:status` | Dashboard showing harness state, metrics, coverage |\n| `/harnesskit:test` | Run tests with failure tracking and error classification |\n| `/harnesskit:lint` | Run configured linter |\n| `/harnesskit:typecheck` | Run TypeScript type checking |\n| `/harnesskit:dev` | Start dev server |\n| `/harnesskit:prd` | Decompose PRD into GitHub issues + feature list |\n| `/harnesskit:worktree` | Harness-aware git worktree for isolated feature work |\n\n> **Note**: `setup` internally calls `init` which handles detection and file generation. The 11th skill (`init`) runs as part of setup and is not called directly.\n\n## Presets\n\nChoose a preset during `/harnesskit:setup` based on your comfort level:\n\n| | Beginner | Intermediate | Advanced |\n|---|---|---|---|\n| **Guardrails** | Strict — confirms before destructive ops, enforces test-before-commit | Moderate — warnings only, skip confirmation for safe ops | Minimal — trust the developer |\n| **Briefing** | Detailed — explains what each hook does | Standard — concise summaries | Silent — no briefing unless errors |\n| **Session hooks** | All enabled + verbose logging | All enabled, compact logging | Selective — only error tracking |\n| **Auto-apply** | Off — always ask before changes | Suggest — show diff, one-click apply | Auto — apply low-risk improvements automatically |\n| **Best for** | New to AI-assisted coding | Regular Claude Code users | Power users, vibe coders |\n\n## Architecture\n\n### Design Principles\n\n- **Marketplace First**: Uses existing Claude Code plugins before creating custom tools. Only customizes when session data shows a gap.\n- **Zero-Token Hooks**: All observation hooks (`session-start`, `guardrails`, `post-edit-lint`, `post-edit-typecheck`, `session-end`) run as bash + jq scripts. They cost zero LLM tokens.\n- **Bible**: A curated set of harness engineering principles, referenced by all skills for consistency.\n\n### Subsystems\n\n| Subsystem | Purpose |\n|-----------|---------|\n| **v2a** — Adaptive Generation | Auto-generates skills, agents, hooks, and rules from session data. Powers the `insights` → `apply` loop. |\n| **v2b** — Advanced Workflows | A/B testing for proposals, PRD decomposition into issues, harness-aware worktree isolation. |\n\n### Plugin Structure\n\n```\nHarnessKit/\n├── .claude-plugin/\n│   ├── plugin.json        # Plugin manifest\n│   └── marketplace.json   # Marketplace catalog\n├── skills/                # 11 skill definitions\n├── agents/                # Orchestrator agent\n├── hooks/                 # Session hooks (bash + jq)\n│   └── hooks.json         # Hook registration (read by Claude Code)\n├── scripts/               # Detection & utility scripts\n├── templates/             # Config templates, presets, bible\n├── tests/                 # 122 tests across 10 suites\n├── docs/                  # Design specs, plans, research\n├── README.md\n├── README.ko.md\n└── LICENSE\n```\n\n## Generated Files\n\nWhen you run `/harnesskit:setup`, HarnessKit generates these files in your project:\n\n| File | Purpose |\n|------|---------|\n| `CLAUDE.md` | Session protocol + framework conventions (composed from templates) |\n| `.claudeignore` | Context exclusion patterns per language |\n| `.claude/settings.json` | Hook registrations (session-start, guardrails, session-end) |\n| `docs/feature_list.json` | Feature tracking (`passes: false` pattern) |\n| `progress/claude-progress.txt` | Session-to-session work continuity |\n| `.harnesskit/config.json` | Preset, schema version, installed plugins, custom toolkit |\n| `.harnesskit/detected.json` | Auto-detected repo properties |\n| `.harnesskit/failures.json` | Error pattern tracking across sessions |\n| `.harnesskit/session-logs/` | Per-session observation data (tool usage, time distribution) |\n| `.harnesskit/bible.md` | Curated harness engineering principles reference |\n\nAll generated files respect your chosen preset.\n\n## How the Improve Loop Works\n\n```\nSessions accumulate data\n        │\n        ▼\n/harnesskit:insights\n  ├── Analyzes: errors, tool usage, time sinks, coverage gaps\n  ├── Proposes: new skills, agents, hooks, rules, or config changes\n  └── Categorizes: skill_creation, hook_creation, review_supplement, ...\n        │\n        ▼\n/harnesskit:apply\n  ├── Shows each proposal with diff preview\n  ├── Options: accept / reject / A/B test\n  └── Executes: generates files, updates config\n        │\n        ▼\nNext sessions benefit from improvements\n```\n\n## Contributing\n\nContributions are welcome.\n\n```bash\n# Clone\ngit clone https://github.com/ice-ice-bear/harnesskit.git\ncd harnesskit\n\n# Run all tests (122 tests across 10 suites)\nfor t in tests/test-*.sh; do bash \"$t\"; done\n```\n\nPlease open an issue before submitting large changes.\n\n## License\n\n[MIT](LICENSE) — Copyright 2026 HarnessKit Contributors\n",
  "bytes": 7838,
  "sha": "c8fd6b5fa2faf9b8ceccbf9d57151e4183fcb2ef51e38e77c41c981f42b6c246",
  "repo_slug": "ice-ice-bear/harnesskit",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_ice_ice_bear_harnesskit_harnesskit_4cb3b898/readme"
}