https://github.com/spikeon/caveman-plan
Epic → Plans → Hunts → Inventory. Large-project agent skill with Boss checkpoints. Built on caveman + superpowers-plan.
https://github.com/spikeon/caveman-plan
agent-skills claude-code claude-skills cursor planning skills
Last synced: 28 days ago
JSON representation
Epic → Plans → Hunts → Inventory. Large-project agent skill with Boss checkpoints. Built on caveman + superpowers-plan.
- Host: GitHub
- URL: https://github.com/spikeon/caveman-plan
- Owner: spikeon
- License: mit
- Created: 2026-05-19T07:58:01.000Z (about 1 month ago)
- Default Branch: main
- Last Pushed: 2026-05-19T08:19:54.000Z (about 1 month ago)
- Last Synced: 2026-05-19T10:53:12.219Z (about 1 month ago)
- Topics: agent-skills, claude-code, claude-skills, cursor, planning, skills
- Size: 10.7 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# caveman-plan
Big project? Epic first. Plans split. Hunt one file. Boss say go. Then next hunt.
Agent skill for **large-project planning and execution**: Epic checklist → multiple hunt plans → one plan per hunt (hard stops) → inventory pass against original Goal. Talks to Boss in [caveman](https://github.com/JuliusBrussee/caveman) voice; plan tasks use [superpowers-plan](https://github.com/obra/superpowers) step shape (files, change, verify).
## What it does
Four phases. Boss checkpoints between each. Planning artifacts live in `.plans/` (gitignored — local only).
| Phase | Output | Boss checkpoint |
|-------|--------|-----------------|
| **Epic** | `.plans/epic_.md` — Goal (verbatim prompt) + Steps (S/M/L hints) | Changes before planning? |
| **Plans** | `.plans/epic_/plan_*.md` — **one plan file per Epic step minimum** + `plans-manifest.md` | Time to go hunting? |
| **Hunts** | Kill tasks in **one** `plan_*.md`, then **end turn** | Follow Ups? → Time for next hunt? |
| **Inventory** | Compare kills to Epic Goal + Steps | Hunting complete (only when Epic satisfied) |
**Hunt rules:** Mark tasks `- [x]` only when sure. Doubt → `## Follow Ups` in same plan, not checked off. **Never** chain multiple plans in one turn. **Never** collapse a multi-step Epic into one plan file.
**Layout:**
```
.plans/
├── epic_auth-refactor.md
└── epic_auth-refactor/
├── plans-manifest.md # step → plan mapping + hunt status
├── plan_jwt-middleware.md
├── plan_migration.md
└── plan_rollout-flag.md
```
## How to invoke
```
/caveman-plan # start Epic from current prompt
/caveman-plan hunt # resume current hunt (one plan only)
/caveman-plan inventory # Phase 4 only
```
**Claude + `/goal` enabled:** set goal at start, Boss says **`NEXT HUNT!`** between plans:
```
/goal Epic file created. All plan files created. Every checkbox in epic and every plan checked off.
```
Natural triggers: "epic plan", "go hunting", **`NEXT HUNT!`**, large multi-file / multi-session work.
**Hunt turn limit:** one plan = max **3 agent turns**. Still open on turn 3 → stop, report blockers, wait for Boss (no silent grind).
## Install
Requires **`caveman`** and **`superpowers-plan`** (or **`writing-plans`** from obra/superpowers) installed — see [Credits](#credits).
**Recommended — [Skills CLI](https://skills.sh/):**
```bash
# Dependencies (global, once)
npx skills add juliusbrussee/caveman --skill caveman -g -y
npx skills add obra/superpowers --skill writing-plans -g -y
# This skill
npx skills add spikeon/caveman-plan -g -y
```
Project-local instead of global: drop `-g`.
Browse: [skills.sh/spikeon/caveman-plan](https://skills.sh/spikeon/caveman-plan)
**Manual — git clone (`~/.agents/skills`):**
```bash
git clone https://github.com/spikeon/caveman-plan.git ~/.agents/skills/caveman-plan
# then copy or symlink skills/caveman-plan/ → ~/.agents/skills/caveman-plan
```
**Cursor project skill (optional):**
```bash
npx skills add spikeon/caveman-plan -y
# or: ln -sfn /path/to/caveman-plan/skills/caveman-plan .cursor/skills/caveman-plan
```
Attach the skill or mention `/caveman-plan` in chat.
## Example flow
**Boss:** "Refactor auth to JWT. No forced logout on deploy."
**Epic** (caveman to Boss): writes `epic_auth-refactor.md` with Goal + 4 Steps → *"Boss: change Epic before planning?"*
**Plans:** 4+ `plan_*.md` files + manifest → *"Boss: time go hunting?"*
**Hunt 1:** only `plan_jwt-middleware.md` → end turn → *"Follow Ups? Next hunt?"*
**Hunt 2+:** Boss says yes → next plan only.
**Inventory:** re-read Goal, check every Epic step → *"Hunting complete."*
## See also
- [`skills/caveman-plan/SKILL.md`](./skills/caveman-plan/SKILL.md) — full agent instructions
- [`skills/caveman-plan/reference.md`](./skills/caveman-plan/reference.md) — templates, manifest examples, anti-patterns
## Credits
This skill **builds on** (install separately):
| Project | Author | Used for |
|---------|--------|----------|
| [**caveman**](https://github.com/JuliusBrussee/caveman) | [Julius Brussee](https://github.com/JuliusBrussee) | Terse caveman voice when talking to Boss during the workflow |
| [**superpowers**](https://github.com/obra/superpowers) (`superpowers-plan`) | [Jesse Vincent (obra)](https://github.com/obra) | Per-task plan format: small steps, files to touch, verify commands |
**caveman-plan** combines those patterns into Epic → Plans → Hunts → Inventory with mandatory Boss checkpoints. It is not affiliated with or maintained by the upstream projects.
## License
MIT — see [LICENSE](./LICENSE).