agents

A monorepo of skills, subagent descriptions, and related tooling for driving local and open-weight LLMs in agentic coding harnesses.

Install

./install.sh

Asks three yes/no questions — one per pack below — and symlinks whichever you pick into ~/.agents/skills/. Point your harness's skills directory there (or symlink your harness's own skills dir to it) and you're done. Re-running is safe; a pre-existing real (non-symlink) directory at the target is backed up first.

Layout

agents/
├── install.sh              unified installer — see Install above
├── CHANGELOG.md
└── skills/
    ├── deepdocs/           Swarm-orchestrated system documentation (docgen wiki) for capable mid-size models
    ├── deepreview/         Swarm-orchestrated deep review for capable mid-size models
    ├── doctools/           Documentation-focused skills
    ├── qol/                Quality-of-life skills for any model or harness
    └── qwen-superpowers/   Superpowers methodology adapted for Qwen3.5-397B-A17B (multi-harness)

More packages — additional skill packs, subagent definitions, and so on — will live alongside. All packs are equal-weight; install.sh treats none as default or optional.

Packages

skills/deepdocs

Swarm-orchestrated documentation of existing systems for capable mid-size models, producing a docgen wiki where every claim is provenance-typed (cited code / attributed human fact / explicit open question) and adversarially verified. Currently: system-documenter — survey, document, re-verify, and harvest modes over a shared human-knowledge base. See its README.

skills/deepreview

Swarm-orchestrated deep review skills for capable mid-size models (DeepSeek V4 Flash class and up) in parallel-subagent harnesses. Currently: feature-review — adversarial, evidence-only review of a feature implementation across several services. See its README.

skills/doctools

Documentation-focused skills: turning messy source material into clean docs, and producing evidence-backed reports on how a codebase works. See its README.

skills/qol

Quality-of-life skills for any model or harness. Currently: html-report — polished, self-contained HTML reports in the docgen house style.

skills/qwen-superpowers

An adaptation of obra/superpowers for Qwen3.5-397B-A17B across multiple harnesses (Qwen Code, Gemini CLI, Claude Code, or any AGENTS.md-aware tool): always-on discipline (AGENTS.md), slash-command entry points, and enforcement hooks that contain a weaker model's mistakes (block destructive shell, run project checks after every edit). The skills install like any other pack via the root install.sh; the hooks/AGENTS.md containment layer is a separate, manual, per-harness step — see its README.

License

MIT. This adaptation of obra/superpowers in skills/qwen-superpowers is independent and not affiliated with or endorsed by the upstream project; see CHANGELOG.md for its version history.