Upstream repo: JuliusBrussee/caveman Type: OpenCode plugin — hooks-based, not CLI installer
| Skill | Source |
|---|---|
| Core caveman behavioral rules | skills/caveman/SKILL.md |
| Commit message formatting | skills/caveman-commit/SKILL.md |
| Review comment format/severity | skills/caveman-review/SKILL.md |
The prompt strings in src/skills/*.ts carry upstream's behavioral rules, intensity levels, examples, auto-clarity exceptions, and boundaries.
| Area | This Plugin | Upstream |
|---|---|---|
| Install | OpenCode plugin via opencode.json + caveman.json config |
CLI installer, flag files, multi-platform hooks |
| caveman-stats | Not included — OpenCode handles token counts natively | Included |
| caveman-compress | Not included — pair with opencode-dcp | Included |
| caveman-help | Not included — OpenCode has /help built-in |
Included |
| caveman-shrink / cavecrew | Not included — MCP middleware, OpenCode not affected | Included |
| Auto-activation | Via hook experimental.chat.system.transform with session state |
Via flag file, env vars, CAVEMAN_DEFAULT_MODE |
| Mode persistence | Per-session in-memory (via state.ts) |
File-based flag files |
| Multi-agent support | No — OpenCode single-agent | 30+ agents |
| Config | caveman.json (project or global) |
~/.config/caveman/config.json + env vars |
To refresh prompts against upstream:
- Fetch upstream SKILL.md files (URLs above)
- Compare rules, examples, auto-clarity, boundaries in
src/skills/*.ts - Update TypeScript prompt strings to reflect upstream changes
- Only adopt behavioral rules — skip installer, stats, compress, help features
- Run
bun testto verify