cavecrew
An agent skill by JuliusBrussee, from JuliusBrussee/caveman. Tags: codebase, developer-tools, multi-agent, review.
What it does
When to delegate to `cavecrew-investigator` (locate code), `cavecrew-builder` (1-2 file edit) or `cavecrew-reviewer` (diff review) instead of working inline or using `Explore`. Their output is compressed, so main context lasts longer.
Install
With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:
npx skills add JuliusBrussee/caveman --skill cavecrew
Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):
git clone --depth 1 https://github.com/JuliusBrussee/caveman
cp -r caveman/plugins/caveman/skills/cavecrew ~/.claude/skills/cavecrew
Safety box score
Grade A, minimal risk: 5 of 100 risk points, rated 2026-10-23. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.
This skill is a routing guide for choosing among subagents; it does not ship executable code, touch files, or call external services. The main thing it changes is agent behavior and context usage: it compresses subagent output and imposes a specific voice/format on those subagents.
| Category | Level | Why |
|---|---|---|
| Code & shell execution | 0 (none) | No scripts, shell commands, code execution, or evaluators are shipped; the only file is `SKILL.md`. |
| Network access | 0 (none) | The skill contains no network calls, download instructions, or remote data submission. |
| Filesystem reach | 0 (none) | It does not read, write, delete, or otherwise reference files outside the skill documentation. |
| Secrets & credentials | 0 (none) | It does not request tokens, passwords, key files, `.env` contents, or browser/CLI sessions. |
| Privilege & persistence | 0 (none) | There is no sudo, global install, service registration, shell startup modification, or OS setting change. |
| Instruction hijack surface | 1 (low) | It imposes a persona/style (`"ultracave voice"`, `"drop caveman → normal English"`) and prescribes output contracts, but it does not tell the agent to ignore safeguards or fetch external instructions. |
| Irreversible actions | 0 (none) | It only recommends which subagent to use; it does not itself push, deploy, send messages, pay, book, or delete anything. |
| Transparency | 0 (none) | Everything provided is plain-text markdown; there are no binaries, blobs, installers, or hidden/runtime-decoded content. |
Before installing
- Read `SKILL.md` and confirm you want a subagent-routing policy, not a tool that changes files or systems.
- Check whether your workflow tolerates the enforced terse output format and `ultracave voice` styling.
- If you rely on detailed prose in reviews, note that `cavecrew-reviewer` returns findings only, not full rationale.
Rated 2026-10-23 by an automated review of 1 file from the skill's repository.
Source
- Repository
- JuliusBrussee/caveman (all skills from this repository)
- Path
- plugins/caveman/skills/cavecrew/SKILL.md
- Branch
- main
- Collection
- caveman
- Updated
- 2026-10-05
Related skills
- code-improver — Runs an autonomous review-and-fix improvement loop over any code target — a skill, plugin, module, or directory — using a reviewer the user names.
- interrogate — Use for "interrogate", "adversarial review", "multi-model review", "challenge this", "stress test this code", "find blind spots", or "tear this apart".
- libafl — Builds custom fuzzers with LibAFL, the modular Rust fuzzing library.
- reflect — Spawn three parallel review subagents over the active transcript, surface learnings, and route each to a concrete edit on an existing skill.
- review-plugin-submission — Audit a Cursor plugin for marketplace readiness. Use when validating manifests, component metadata, discovery paths, and submission quality before publishing.
- stitch-loop — Teaches agents to iteratively build websites using Stitch with an autonomous baton-passing loop pattern