{
  "markdown": "# 🦫 Beaver\n\n> Skills for system design, software architecture, and specs — structure before code.\n\nBeavers are nature's engineers. They read the current before they build, raise a dam\nthat bends the whole river, and shape a lodge log by log — nothing placed that the\nstructure doesn't need. Beaver brings that instinct to software: **survey the terrain,\nframe the architecture, and craft the code — deliberately, structure before code.**\n\nIt's a collection of [Agent Skills](https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills)\nthat follow the open `SKILL.md` standard, so they work across skill-aware tools —\nClaude Code, Codex CLI, Gemini CLI, Cursor, and more.\n\n## What's inside\n\n| Skill | Stage | What it does |\n|---|---|---|\n| [`defining-products`](skills/defining-products) | 🧭 Scout the valley | Idea → decision-grade PRD: users and problems, a prioritized feature list with stable requirement IDs, product-level acceptance criteria, success metrics, milestones, and explicit non-goals. |\n| [`designing-systems`](skills/designing-systems) | 🌊 Survey the current | Idea → decision-grade high-level design: constraint register, building-block selections, component inventory, data/ML flows, quality attributes, risks, SMART criteria, next-phase spec backlog. Works for software, data/ML/AI, IoT/edge, and cyber-physical systems. |\n| [`architecting-software`](skills/architecting-software) | 🪵 Raise the dam | Design or review a service against **Clean Architecture** — domain independent of frameworks, DB, and providers. |\n| [`specifying-features`](skills/specifying-features) | 📏 Mark the timber | Feature → implementable spec: scope, contract deltas, data changes, edge cases, executable acceptance criteria, and an explicit Definition of Done. |\n| [`crafting-code`](skills/crafting-code) | 🏠 Shape the lodge | A coding standard: a blocking process gate, four quality pillars, and the ponytail simplicity ladder. |\n| [`verifying-completion`](skills/verifying-completion) | 💧 Hold water | Build → completion report: every requirement traced to executed evidence, gaps named by failure mode, and an earned done / not-done verdict. |\n\nThey chain — **defining-products** → **designing-systems** → **architecting-software**\n→ **specifying-features** → **crafting-code** → **verifying-completion** — or stand\nalone. Coupling is loose by design: each skill consumes upstream artifacts when it\nfinds them in the project and works from the request alone when it doesn't.\n\n## Install\n\nThe skills follow the open `SKILL.md` standard, so they work across skill-aware tools.\nPick whichever path fits your setup.\n\n### Claude Code — marketplace\n\n```\n/plugin marketplace add dlombard/beaver\n/plugin install beaver@beaver\n```\n\nInstalls all six skills; each triggers on its own by description. Update later with\n`/plugin marketplace update beaver`.\n\n### Gemini CLI — extension\n\nThis repo is also a Gemini CLI extension (`gemini-extension.json`); Gemini\nauto-discovers the bundled `skills/`:\n\n```bash\ngemini extensions install https://github.com/dlombard/beaver\n```\n\n(Check `gemini extensions --help` for the current install syntax.)\n\n### Codex, Cursor & other tools — the `.agents/skills/` convention\n\nMost non-Claude tools read skills from **`.agents/skills/`**, an emerging cross-tool\nstandard (Codex CLI and Gemini CLI both scan it; `.agents/skills/` takes precedence in\nGemini). Just drop the skills there:\n\n```bash\ngit clone https://github.com/dlombard/beaver.git\ncp -R beaver/skills/* ~/.agents/skills/          # user-level, shared across tools\n# …or per project:\ncp -R beaver/skills/* <your-repo>/.agents/skills/\n```\n\nTool-specific homes also work (`~/.codex/skills/`, `~/.gemini/skills/`,\n`~/.claude/skills/`); a common setup is one source of truth symlinked into the others.\n\nbeaver is also packaged as a **Codex plugin** (`.codex-plugin/plugin.json`, listed in\n`.agents/plugins/marketplace.json`) for a marketplace-style install on Codex, in addition\nto the `.agents/skills/` drop-in above.\n\n*Skills directories evolve per tool — check each tool's current Agent Skills docs.*\n\n## Examples & evaluations\n\nEvery skill ships a runnable example with an expected outcome and a **pass/fail\ngrading rubric** in [`examples/`](examples/). Run one in a fresh session and grade the\nresult against its rubric — that's how we keep the skills honest.\n\nThe deeper **A/B harness** lives in [`evals/`](evals/): it runs each case with and\nwithout the skill (and in a trigger mode that checks the skill fires on its own,\nscored from the transcript), so a skill has to show **uplift over a no-skill\nbaseline** to earn its place. Graded results are in [`evals/GRADING.md`](evals/GRADING.md);\nthe adversarial review in [`evals/REVIEW.md`](evals/REVIEW.md).\n\n## Contributing\n\nNew skills welcome — one lodge at a time. See [CONTRIBUTING.md](CONTRIBUTING.md).\n\n## License\n\nMIT — see [LICENSE](LICENSE). The `crafting-code` skill embeds the MIT-licensed\n[ponytail](https://github.com/DietrichGebert/ponytail) simplicity ladder\n(© Dietrich Gebert); its license travels with the skill.\n",
  "bytes": 5106,
  "sha": "a53df10e01188fef64a8ae10aad785c67d0c9634df24f2ee8aaa892e017b7032",
  "repo_slug": "dlombard/beaver",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_dlombard_beaver_e3e8aa59/readme"
}