Cabinet is an intentionally small, curated working library of proven operator workflows for AI-assisted software work.
It is built like a field manual, not a broad platform: direct, calm, reusable, and kept deliberately narrow so each skill has to earn its place. The public library is small on purpose. Every included skill solves a repeatable operational problem without leaning on one private repo, one client, or one internal process.
A few concrete examples:
- If you inherit an unfamiliar repository and need to decide whether it is safe to inspect, sandbox, or trust with real data, use
repo-audit. Outcome: a conservative plain-language trust assessment and a clear next step. - If you need to pause an agent session and continue later without reconstructing the whole project, use
handoffandresume-from-handoff. Outcome: continuation-grade context transfer instead of restating everything from scratch.
repo-audithelps you decide whether a repository is safe to read, clone, run in a sandbox, or trust with real data.- It is optimized for fast, conservative risk assessment with plain-language output.
handoffwrites a continuation-grade markdown handoff for another agent.resume-from-handoffloads that handoff, verifies the live state, and continues without making the user restate the whole project.- These live under
skills/handoff-protocol/as separate skills on disk, but they form one operational workflow.
end-of-daycloses a work interval cleanly: review merge readiness, summarize shipped work, sync stale docs, capture next steps, and prepare a session for sign-off or handoff.- In Cabinet,
end of daymeans since the last trustworthy checkpoint, not strictly the clock.
screenshot-cleanupcleans up a screenshot or image dump on macOS by generating a temporary OCR-assisted helper, sorting files into category folders, and renaming them to readable slugs.- It is intentionally concrete, explicitly macOS-only, and explicit about local side effects.
A lot of useful skills are trapped inside one person's environment. Cabinet is an attempt to publish a small public working set of the good ones without sanding off the operational detail that makes them useful.
The quality bar is simple:
- reusable outside one private codebase
- concrete enough to execute
- compact enough to maintain
- honest about verification and uncertainty
Cabinet adds a new top-level skill only when it clears all of these bars:
- it solves a narrow, repeatable operator problem
- it states platform assumptions plainly when they matter
- it declares meaningful local side effects and safety boundaries up front
- it stays Markdown-first; generated helpers are fine, committed executables are not the default
- it improves the library more than it broadens it
cabinet/
├── README.md
├── LICENSE
├── CONTRIBUTING.md
├── CODE_OF_CONDUCT.md
├── chat-session-summaries/
├── council-decisions/
├── docs/
│ ├── adding-a-skill.md
│ └── roadmap.md
└── skills/
├── repo-audit/
│ ├── SKILL.md
│ └── references/
├── handoff-protocol/
│ ├── handoff/
│ │ └── SKILL.md
│ └── resume-from-handoff/
│ └── SKILL.md
├── end-of-day/
│ └── SKILL.md
├── screenshot-cleanup/
│ └── SKILL.md
└── _template/
└── SKILL.md
- Copy the skill folder you want into your own skills directory or prompt system.
- Keep any local
references/directory intact when a skill depends on companion files. - Adapt outer framing only when necessary; preserve the core workflow unless you have a reason to change it.
- Prefer using the skills as operational building blocks, not as decorative prompt text.
Cabinet is Markdown-first. There is no package manager, site generator, or release automation in the first cut.
This is the current public working set. It is intentionally selective, not a completeness claim.
repo-audithandoffresume-from-handoffend-of-dayscreenshot-cleanup_templateas the starter shape for future additions
- See
docs/adding-a-skill.mdfor the contribution standard. - See
docs/roadmap.mdfor the near-term direction and current non-goals.
Cabinet is released under the MIT License.
