{
  "markdown": "# Skill Hunter\n\n![Skill Hunter — before you build, scan the ecosystem](./assets/og-image.png)\n\n**Stop your AI agent from reinventing the wheel.**\n\nSkill Hunter is a pre-execution layer for coding and automation agents. It makes agents check for existing skills, MCP servers, CLIs, packages, APIs, templates, repos, and workflows before building from scratch.\n\n> Your agent should check the toolbox before building a hammer.\n\n## The Problem\n\nAgents are good at writing code, sometimes too good. They will happily generate a custom scraper, parser, integration, or workflow even when a safer maintained tool already exists.\n\nSkill Hunter adds the missing pause: understand the task, search the ecosystem, compare options, score risk, ask for approval when needed, then reuse or build minimally.\n\n## Before / After\n\n**Before**\n\nUser asks: \"Parse these PDFs and extract invoices.\"\n\nAgent writes a custom parser immediately.\n\n**After Skill Hunter**\n\nAgent checks PDF libraries, OCR tools, invoice extraction APIs, existing repo utilities, and asks whether accuracy, cost, privacy, or offline processing matters before writing code.\n\n## What It Does\n\n1. Classifies the request.\n2. Checks whether reusable solutions likely exist.\n3. Compares candidates by fit, maintenance, permissions, security, docs, license, and effort.\n4. Recommends reuse, adaptation, minimal build, custom build, or asking the user.\n5. Gates risky installs, credentials, external services, posting, and destructive actions on approval.\n\n## Install\n\n### Claude Code plugin\n\n```text\n/plugin marketplace add mturac/skill-hunter\n/plugin install skill-hunter@skill-hunter\n```\n\n### Codex CLI plugin\n\n```bash\ncodex plugin marketplace add mturac/skill-hunter\n```\n\nThen enable it in `~/.codex/config.toml`:\n\n```toml\n[plugins.\"skill-hunter@skill-hunter\"]\nenabled = true\n```\n\n### OpenClaw / ClawHub\n\n```bash\nopenclaw skills install openclaw-skill-hunter\n# or\nclawhub install openclaw-skill-hunter\n```\n\nPublished at [clawhub.com/mturac/openclaw-skill-hunter](https://clawhub.com/mturac/openclaw-skill-hunter).\n\n### `npx skills`\n\n```bash\nnpx skills add mturac/skill-hunter@skill_hunter -g -y\n```\n\n### Local clone\n\n```bash\ngit clone https://github.com/mturac/skill-hunter.git\ncd skill-hunter\n./install.sh all\n./install.sh status\n```\n\n## Supported Runtimes\n\n| Runtime | Install surface | Hook support |\n|---|---|---|\n| Claude Code | plugin or `./install.sh claude` | interactive `UserPromptSubmit` |\n| Codex CLI | plugin or `./install.sh codex` | interactive `UserPromptSubmit` |\n| OpenClaw | `./install.sh openclaw` or ClawHub | skill only |\n| Hermes / agentskills.io | `npx skills` or `./install.sh hermes` | skill only |\n| Cursor | `./install.sh cursor` | rule file |\n\nThe hook only fires in interactive Claude/Codex sessions. Headless modes rely on skill matching.\n\n## Provider Status\n\nSkill Hunter already ships through real provider surfaces: Claude Code plugin metadata, Codex CLI plugin metadata, OpenClaw/ClawHub install paths, `npx skills`, and local installer targets.\n\nThe future work is automated discovery provider adapters: code that queries ClawHub, GitHub, MCP registries, npm, PyPI, or local docs and returns scored candidates through a CLI/API.\n\n## Output Shape\n\n```text\nSkill Discovery Pass:\n- Goal:\n- Existing options checked:\n- Best option:\n- Decision:\n- Risk:\n- Next action:\n```\n\nDecision values: `USE_EXISTING`, `ADAPT_EXISTING`, `BUILD_MINIMAL`, `BUILD_CUSTOM`, `ASK_USER`, `AVOID`.\n\n## Trust Score\n\nThe planned Trust Score turns discovery into a transparent decision model:\n\n- Relevance, 0-25\n- Maintenance, 0-20\n- Adoption, 0-15\n- Security, 0-25\n- Documentation, 0-10\n- Fit cost, 0-5\n\nSee [docs/TRUST_SCORE.md](./docs/TRUST_SCORE.md).\n\n## Security Model\n\nSkill Hunter never silently installs or executes unknown tools. Approval is required for risky installs, credentials, paid/external services, email/calendar/social posting, networked execution, destructive commands, or writes outside the repo.\n\nSee [docs/SECURITY_MODEL.md](./docs/SECURITY_MODEL.md).\n\n## Examples\n\nSee [examples/](./examples) for PDF conversion, scraping, trivial-skip behavior, custom parser decisions, and credential-scope warnings.\n\n## Benchmarks\n\nBenchmarks live in [benchmarks/tasks.yaml](./benchmarks/tasks.yaml). They compare baseline agent behavior against Skill Hunter guided behavior across scraping, PDF parsing, PR review, and approval-gated social posting.\n\n## Roadmap\n\n- P0: centralized prompts, docs, benchmark examples, install/version consistency.\n- P1: Trust Score implementation, security audit model, CLI skeleton, local audit mode, JSON output.\n- P2: automated ClawHub/GitHub/MCP/npm/PyPI discovery adapters, GitHub Action, demo GIF, public benchmark results.\n\nSee [docs/ROADMAP.md](./docs/ROADMAP.md).\n\n## Known Limitations\n\n- Distribution provider surfaces are real; automated discovery adapters are not implemented yet.\n- Benchmark results are illustrative until a runner exists.\n- OpenClaw currently installs a skill copy, not a hook.\n- Skill Hunter reduces unsafe behavior, but does not replace human security review.\n\n## Contributing\n\nGood first issues: provider stubs, Trust Score implementation, benchmark runner, adapter consistency, JSON output mode, and release checklist.\n\n## License\n\nMIT — see [LICENSE](./LICENSE).\n\n---\n\n## Part of [mturac/tools](https://github.com/mturac/tools)\n\nThis project is part of an open-source toolkit for AI-augmented engineering — Claude Code plugins, MCP servers, security scanners, schedulers, and dev-productivity utilities. See the [hub](https://github.com/mturac/tools) for the full list.\n\nInstall every Claude Code plugin from one place:\n\n```text\n/plugin marketplace add mturac/claude-plugin-marketplace\n/plugin install skill-hunter\n```\n\n",
  "bytes": 5757,
  "sha": "62657e6c0852df81004a390c1b7ec0039c8c36553c58824fd126353b79a0df08",
  "repo_slug": "mturac/skill-hunter",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_mturac_skill_hunter_skill_hunter_1ab947a4/readme"
}