Skills

My collection of AI agent skills — to work with safety and quality as a real software engineer.

---

Available skills

Skill	What it does
write-great-agentsmd	Creates, updates, and audits AGENTS.md — the policy that orchestrates the agent in the repo. Branches: bootstrap (from scratch), update (sync after changes), lint (health audit).
commit-message	Generates Conventional Commits messages from the real diff. Fires only when asked; never force-pushes, skips hooks, or commits secrets.
code-review-plus	Code review across 5 dimensions (correctness, security, architecture, quality, performance) with parallel subagents and anti-false-positive checks. Prioritized report (P0–P3), calibrated to avoid inflated severity.
make-docs	Generates and maintains architecture docs and behavioral specs in docs/. Branches: explore (survey → propose → generate), update (commits since the Updated on stamp), adr (decision record).

Each skill has a SKILL.md (source of truth for the agent).

Some include a human-facing README.md, catalogues (PATTERNS.md), or templates under references/.

---

Quickstart

1. Install via the skills.sh script:

   • All skills:

   npx skills add diegoos/skills

   • A specific skill:

   # e.g. install the write-great-agentsmd skill
   npx skills add diegoos/skills --skill write-great-agentsmd

2. Done — you can use the installed skill(s).

---

Structure

skills/
├── write-great-agentsmd/   # AGENTS.md — bootstrap / update / lint
│   ├── SKILL.md
│   ├── PATTERNS.md
│   └── README.md
├── commit-message/         # Conventional Commits
│   └── SKILL.md
├── code-review-plus/       # Multi-perspective review
│   └── SKILL.md
└── make-docs/              # Architecture + observable specs
    ├── SKILL.md
    ├── README.md
    └── references/         # Templates (architecture, ADR, specs, …)

.opencode/                  # Optional — Custom OpenCode agents
└── agents/
    ├── ask.md              # Read-only agent
    └── reviewer.md         # Single-pass review

---

Custom OpenCode agents

Custom agents for OpenCode can live under .opencode/agents/.

They are not part of the skills contract.

Agent	Role
ask	Read-only persona: conversation and code analysis. No edits, no bash, no subagents.
reviewer	Single-pass, high-signal review persona. Complements code-review-plus (which does the deep parallel review).

---

How to use

In some cases the agent loads the skill from intent — you don't need to name it:

"Create an AGENTS.md for this project"           → write-great-agentsmd (bootstrap)
"Generate docs for this codebase"                → make-docs (explore)
"Update the docs after these changes"            → make-docs (update)
"Record the decision to use Postgres"            → make-docs (adr)

Or invoke by skill name/command (e.g. /make-docs explore), depending on the harness.