Custom skills, agents, and commands for opencode (and other coding agents).
This repository contains personalized extensions that enhance coding agent capabilities:
- Skills — Specialized workflows for common tasks
- Agents — Reusable agent definitions with specific tool access
- Commands — Quick actions triggered with arguments
- Scripts — Maintenance helpers (e.g. skill validation)
| Skill | Trigger | Description |
|---|---|---|
| git-commit | "commit", "git commit", "create a commit" | Storytelling-focused Conventional Commits with human-in-the-loop "why" gathering. |
| changelog-generator | "create changelog", "release notes" | Turns commit history into user-friendly changelog entries. |
| karpathy-guidelines | Writing/reviewing/refactoring code | Guardrails to reduce common LLM coding mistakes: surgical changes, simplicity first, verifiable success. |
| debug-loop | Bug hunting, especially flaky/intermittent | Reproduce → isolate → hypothesize → failing test → fix → verify. |
| pr-review | Opening or reviewing a PR | Pre-PR checklist + structured review framework. Pairs with code-reviewer agent. |
| spec-to-plan | "create a spec", "plan this feature" | 5-phase workflow: spec → clarify → markdown spec → todo → plan. |
| python-tooling | Python project setup, migrating from pip/black/mypy | Modern stack: uv, ruff, ty. |
| python-testing-patterns | Writing Python tests | pytest, fixtures, mocking, parametrize, async, coverage. Slim index + topic references. |
| typescript-tooling | TypeScript project setup or modernization | Bun + Biome + tsc. Includes monorepo guidance. |
| deep-performance-audit | "performance audit", "optimize codebase" | Hyper-intensively investigate the codebase to identify gross inefficiencies and propose isomorphic improvements. |
| deep-project-primer | "project primer", "initialize project" | Initialization instructions for any project. Investigates code to understand architecture and purpose. |
| idea-wizard | "generate ideas", "improve project" | Generate, evaluate, and implement ideas to improve the project. Generates 30 ideas, filters and plans the top ones. |
| readme-reviser | "update readme", "revise docs" | Update the README and other documentation to reflect all of the recent changes to the project. |
| Agent | Tool access | Purpose |
|---|---|---|
| code-reviewer | read-only + git diff/log + todowrite/question | Reviews recent changes; outputs Critical / Warnings / Suggestions. |
| refactor | read + edit (with ask) + todowrite/question |
Cautious behavior-preserving refactors. Embeds karpathy-guidelines. |
| planner | read-only + todowrite/question + extended bash/br read | Planning agent with strong clarifying-question discipline. Custom personality on top of opencode's built-in plan mode. |
opencode also ships built-in build and plan agents — referenced by some commands below.
| Command | Agent | Purpose |
|---|---|---|
| test | build (built-in) |
Run pytest with coverage; prefers uv run pytest, falls back to python3 -m pytest. |
| clean-init | build (built-in) |
Analyze codebase and write/update AGENTS.md. |
| bug-hunter | general |
Randomly explore code to find and fix bugs. |
| code-reorganizer | planner |
Propose a reorganization plan for scattered code files. |
| de-slopify | refactor |
Remove AI slop style writing from text. |
| Script | Purpose |
|---|---|
| validate-skills.sh | Lint every skills/*/SKILL.md for required frontmatter (name, description) and verify name matches the directory. Exits non-zero on failure. |
Run the skill validator any time you add or modify a skill:
./scripts/validate-skills.shOutput:
ok [changelog-generator]
ok [debug-loop]
...
Checked: 13 Errors: 0
These configurations are designed for opencode. Place this repository at ~/.config/opencode/ or symlink it:
ln -s /path/to/ocskillz ~/.config/opencodeMIT. All bundled skills declare license: MIT in their frontmatter for consistency.