cavecrew Skill for Claude: When to Use Caveman Subagents
cavecrew tells Claude when to hand work to three caveman subagents that return short results, so your main context lasts longer.
- Installs
- 366.1K
- on skills.sh
- Stars
- 116.7K
- on GitHub
- Files
- 1
- besides SKILL.md
- Updated
- Nov 2026
- last repo commit
cavecrew is a decision guide, not a slash command. It tells Claude when to delegate to cavecrew-investigator (find code), cavecrew-builder (edit one or two files) or cavecrew-reviewer (review a diff) instead of working inline or using the built-in Explore agent. Their answers come back compressed, as path:line tables and one-line findings.
The skill only describes when to use the subagents. The subagents themselves come with the caveman package, so install the plugin, not just this folder. It triggers on phrases like "delegate to subagent", "use cavecrew" and "save context".
Good for
- Claude Code users who run many subagent calls and fill their context
- Developers who want find, fix and review steps with short, structured output
- People who already use caveman and want it in subagents
Install cavecrew
Claude Code
- 1
Copy this prompt.
Install the cavecrew skill in Claude Code. Run this command: /plugin marketplace add JuliusBrussee/caveman /plugin install caveman@caveman Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Pick the plugin. The cavecrew skill describes three subagents (cavecrew-investigator, cavecrew-builder, cavecrew-reviewer), and the caveman README lists cavecrew as part of the package. This folder holds only the guide (SKILL.md and a README).
From the project README. Source.
What's in the skill
When it loads
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.
Sections in SKILL.md
- When to use cavecrew vs alternatives
- Why this exists (the real win)
- Output contracts
- Chaining patterns
- What NOT to do
- Auto-clarity (inherited)
Key facts
- License
- Apache-2.0
- Installs on skills.sh
- 366,093 as of 2026-11-02
- GitHub stars
- 116,702 for the whole juliusbrussee/caveman repo
- skills.sh audits
- Gen Agent Trust Hub pass, Socket pass, Snyk pass
- Last commit to the repo
- 2026-11-01
What it includes
Finds where code is defined or used; read-only
Answers "where is X defined" or "what calls Y" as a path:line table, or "No match."
Try: “Use cavecrew to find every caller of parseConfig.”
Read ·
cavecrew-investigatorMakes a surgical edit in 1 to 2 files
Returns the changed lines and a verified: line. It answers too-big, needs-confirm, ambiguous or regressed when it should not proceed.
Write ·
cavecrew-builderReviews a diff, branch or file for bugs
Returns one line per finding sorted by file and line, then totals by severity, or "No issues."
Read ·
cavecrew-reviewerTable of tasks and the right agent
Prose commentary: use vanilla Explore or Code Reviewer. A 3+ file refactor: main thread. A one-line answer you know: no subagent.
Read ·
When to use whichLocate, fix, verify; parallel scouts; single-shot edit
Investigator finds sites, builder edits 1 to 2 of them, reviewer audits the diff. Broad searches spawn 2 to 3 investigators at once.
Read ·
Chaining patternsDo not use the builder when you do not know the file
Warns against chaining investigator to builder for a 5-file refactor, and against asking the reviewer for general feedback.
Read ·
What not to do
From the official docs
Permissions
Low risk
The skill is only a guide, but one of the subagents it points to, the builder, edits files. Check its changes with the reviewer agent or git diff.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
The builder subagent edits your files
It is limited to 1 to 2 files and verifies by re-reading, but you should still read the diff.
Compressed output can be cryptic
SKILL.md warns the results can be terse to the point of cryptic. If a human will read them, ask Claude to paraphrase.
Technical details
| Scopes |
|
|---|
Questions
What does the cavecrew skill do?
It tells Claude when to delegate to three caveman subagents, one that finds code, one that makes small edits and one that reviews diffs. Their results come back short, which saves main-context tokens.
How do I install cavecrew?
Use the install box at the top of this page and pick the caveman plugin. The skill is a guide to the subagents, and the caveman package is what provides them.
Is cavecrew a slash command?
No. The bundled README says it is a decision guide that activates when the conversation mentions delegation, for example "use cavecrew" or "spawn investigator".
Which model do the subagents use?
Per the bundled README, cavecrew-reviewer and cavecrew-investigator pin model: haiku in their settings, and cavecrew-builder uses your session default. The README says you can override each with environment variables.
Can the builder agent change my files?
Yes, up to two files. SKILL.md says it makes surgical edits, refuses bigger scopes and re-reads its edit to verify. Review the diff, or run cavecrew-reviewer on it.
cavecrew vs Claude's built-in Explore agent?
Explore returns prose and suggestions. cavecrew-investigator returns a compact path:line table. SKILL.md says to use Explore when you want commentary and cavecrew when you want output in about a third of the tokens.
Is cavecrew safe?
The skill file is instructions only, but the builder subagent edits files. On skills.sh the audits read Gen Agent Trust Hub pass, Socket pass and Snyk pass. Review what the builder changes.
Information
updated Nov 2, 2026
- Repository
- juliusbrussee/caveman
- Folder
skills/cavecrew
Sources
Docs, issue trackers and discussions this page is built from.
Show all 4Hide
- caveman README · github.com
- cavecrew README · github.com
- cavecrew SKILL.md · github.com
- skills.sh listing and audits · skills.sh