beaver
Skills for system design, software architecture, and specs — structure before code.
Open source Open in the app JSON README (API)
About
Skills for system design, software architecture, and specs — structure before code.
Details
- Kind
- Plugins
- Topic
- No topic detected
- Publisher
- dlombard
- Origin
- gemini
- Category
- ferramentas
- Version
- 0.4.0
- Last push
- 2026-07-12T00:27:13Z
- Repository state
- ativo
- Language
- Shell
- License
- NOASSERTION
- Added
- 2026-08-30 14:13:39
- Updated
- 2026-08-30 14:13:39
- Origin id
dlombard/beaver
README
# 🦫 Beaver > Skills for system design, software architecture, and specs — structure before code. Beavers are nature's engineers. They read the current before they build, raise a dam that bends the whole river, and shape a lodge log by log — nothing placed that the structure doesn't need. Beaver brings that instinct to software: **survey the terrain, frame the architecture, and craft the code — deliberately, structure before code.** It's a collection of [Agent Skills](https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills) that follow the open `SKILL.md` standard, so they work across skill-aware tools — Claude Code, Codex CLI, Gemini CLI, Cursor, and more. ## What's inside | Skill | Stage | What it does | |---|---|---| | [`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. | | [`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. | | [`architecting-software`](skills/architecting-software) | 🪵 Raise the dam | Design or review a service against **Clean Architecture** — domain independent of frameworks, DB, and providers. | | [`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. | | [`crafting-code`](skills/crafting-code) | 🏠 Shape the lodge | A coding standard: a blocking process gate, four quality pillars, and the ponytail simplicity ladder. | | [`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. | They chain — **defining-products** → **designing-systems** → **architecting-software** → **specifying-features** → **crafting-code** → **verifying-completion** — or stand alone. Coupling is loose by design: each skill consumes upstream artifacts when it finds them in the project and works from the request alone when it doesn't. ## Install The skills follow the open `SKILL.md` standard, so they work across skill-aware tools. Pick whichever path fits your setup. ### Claude Code — marketplace ``` /plugin marketplace add dlombard/beaver /plugin install beaver@beaver ``` Installs all six skills; each triggers on its own by description. Update later with `/plugin marketplace update beaver`. ### Gemini CLI — extension This repo is also a Gemini CLI extension (`gemini-extension.json`); Gemini auto-discovers the bundled `skills/`: ```bash gemini extensions install https://github.com/dlombard/beaver ``` (Check `gemini extensions --help` for the current install syntax.) ### Codex, Cursor & other tools — the `.agents/skills/` convention Most non-Claude tools read skills from **`.agents/skills/`**, an emerging cross-tool standard (Codex CLI and Gemini CLI both scan it; `.agents/skills/` takes precedence in Gemini). Just drop the skills there: ```bash git clone https://github.com/dlombard/beaver.git cp -R beaver/skills/* ~/.agents/skills/ # user-level, shared across tools # …or per project: cp -R beaver/skills/* <your-repo>/.agents/skills/ ``` Tool-specific homes also work (`~/.codex/skills/`, `~/.gemini/skills/`, `~/.claude/skills/`); a common setup is one source of truth symlinked into the others. beaver is also packaged as a **Codex plugin** (`.codex-plugin/plugin.json`, listed in `.agents/plugins/marketplace.json`) for a marketplace-style install on Codex, in addition to the `.agents/skills/` drop-in above. *Skills directories evolve per tool — check each tool's current Agent Skills docs.* ## Examples & evaluations Every skill ships a runnable example with an expected outcome and a **pass/fail grading rubric** in [`examples/`](examples/). Run one in a fresh session and grade the result against its rubric — that's how we keep the skills honest. The deeper **A/B harness** lives in [`evals/`](evals/): it runs each case with and without the skill (and in a trigger mode that checks the skill fires on its own, scored from the transcript), so a skill has to show **uplift over a no-skill baseline** to earn its place. Graded results are in [`evals/GRADING.md`](evals/GRADING.md); the adversarial review in [`evals/REVIEW.md`](evals/REVIEW.md). ## Contributing New skills welcome — one lodge at a time. See [CONTRIBUTING.md](CONTRIBUTING.md). ## License MIT — see [LICENSE](LICENSE). The `crafting-code` skill embeds the MIT-licensed [ponytail](https://github.com/DietrichGebert/ponytail) simplicity ladder (© Dietrich Gebert); its license travels with the skill.