cavecrew

Use it when: the agent hands many searches, small edits or reviews to subagents and its main context keeps filling up
It does: tells the agent when to use three terse subagents, a code finder, a one-or-two-file editor and a diff reviewer, instead of working inline
You get: subagent reports in short fixed formats, such as file:line lists, so each handoff costs the main context far fewer tokens
Needs: the Caveman plugin installed: the three subagents come with it, not with this skill
Unlike default subagents: such as Claude Code's built-in Explore search agent, which reply in prose; those stay the choice when you want suggestions or architecture advice

Skill 7 of 12 in the Caveman bundle · Before it: caveman-help · After it: investigate-first

Only works inside its bundle: Delegates to three subagents that come with the Caveman plugin, not with the skill. Install Caveman instead.

TLDR

Caveman is a skill set and plugin that makes agents terse to save tokens. Subagent results land in the main context word for word, so a search that returns 2,000 tokens of prose costs 2,000 tokens every time. cavecrew is a guide to three Caveman subagents that answer in clipped fragments. cavecrew-investigator finds where something is defined or used and returns a file:line list. cavecrew-builder makes a small edit in one or two files whose location is already known, and reports what changed and whether a re-read confirmed it, or turns the task down as too big. cavecrew-reviewer reviews a diff and returns one line per finding, with a tally by severity. The usual chain is locate, fix, review, and broad searches can run two or three investigators at once. New features, edits across three or more files, and reviews that need rationale stay in the main conversation or go to the default agents. The subagents switch to plain English for security warnings and irreversible actions.

Summarizes the skill at bad48a0.

npx skills add https://github.com/juliusbrussee/caveman --skill cavecrew

skills.sh page · Source on GitHub · All skills from juliusbrussee/caveman

Installs: 313,658 on skills.sh, Sep 21, 2026
Repo stars: 106,976 juliusbrussee/caveman
First seen: May 1, 2026 on skills.sh

Weekly installs

13.5K

-9% over 8 weeks · week to Sep 21

Weekly values

Week to	Installs
Aug 3	14,891
Aug 10	15,199
Aug 17	10,221
Aug 24	9,838
Aug 31	9,829
Sep 7	11,870
Sep 14	12,939
Sep 21	13,540

Category	category:output-style category:agents category:software-engineering Change how the agent replies: shorter, clearer, fewer tokens. Build agents, orchestrate workers and give agents memory. Everything for building software: planning, coding, debugging, design, frameworks, cloud, agents and skills.
Type	type:workflow A multi-step process the agent follows: interview, debug, run a TDD loop, review.
Effect	effect:none Adds no side effects of its own. It shapes how the agent thinks, writes or reviews. Guidelines and routers are always effect:none, even when the task they support edits code.
Needs	needs:subagents Relies on the agent being able to spawn subagents.
Publisher	by:community Everyone else. Unverified vendor claims default to community.
Topics	multi-agent token-saving

SKILL.md

From juliusbrussee/caveman at bad48a0. Checked daily, last changed Oct 20, 2026.

Frontmatter

name: cavecrew
description: >
  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.

Cavecrew = three subagent presets that answer in ultracave voice (tool runs: one status line in, one out; nothing between routine calls). Same job as Anthropic defaults (Explore, edit-style agents, reviewer); difference is the tool-result they return is compressed, so main context shrinks per delegation.

When to use cavecrew vs alternatives

Task	Use
"Where is X defined / what calls Y / list uses of Z"	`cavecrew-investigator`
Same but you also want suggestions/architecture commentary	`Explore` (vanilla)
Surgical edit, ≤2 files, scope obvious	`cavecrew-builder`
New feature / 3+ files / cross-cutting refactor	Main thread, or `feature-dev:code-architect` if installed
Review diff, branch, or file for bugs	`cavecrew-reviewer`
Deep code review with rationale + alternatives	`Code Reviewer` (vanilla) if installed, else main thread
One-line answer you already know	Main thread, no subagent

Rule of thumb: if you'd want the subagent's output in 1/3 the tokens, pick cavecrew. If you'd want prose, pick vanilla.

Why this exists (the real win)

Subagent tool results get injected into main context verbatim. A vanilla Explore that returns 2k tokens of prose costs 2k tokens of main-context budget every time. The same finding from cavecrew-investigator comes back as a compressed path:line table instead. Across 20 delegations in one session that's the difference between context exhaustion and finishing the task.

Output contracts

What main thread can rely on per agent:

cavecrew-investigator

<Header>:
- path:line — `symbol` — short note
totals: <counts>.

Or No match. Always file-path-first, line-number-attached, backticked symbols. Safe to grep with path:\d+.

cavecrew-builder

<path:line-range> — <change ≤10 words>.
verified: <re-read OK | mismatch @ path:line>.

Or one of: too-big. / needs-confirm. / ambiguous. / regressed. (terminal first token).

cavecrew-reviewer

path:line: <emoji> <severity>: <problem>. <fix>.
totals: N🔴 N🟡 N🔵 N❓

Or No issues. Findings sorted file → line ascending.

Chaining patterns

Locate → fix → verify (most common):

cavecrew-investigator returns site list.
Main thread picks 1-2 sites, hands paths to cavecrew-builder.
cavecrew-reviewer audits the diff.

Parallel scout (when investigation is broad): Spawn 2-3 cavecrew-investigator calls in one message (different angles: defs vs callers vs tests). Aggregate in main thread.

Single-shot edit (when site is already known): Skip investigator. Hand exact path:line to cavecrew-builder directly.

What NOT to do

Don't use cavecrew-builder when you don't already know the file. Spawn investigator first or main thread will eat tokens passing context.
Don't chain cavecrew-investigator → cavecrew-builder for a 5-file refactor. Builder will return too-big. and you'll have wasted a turn.
Don't ask cavecrew-reviewer for "general feedback" — it returns findings only, no architecture opinions. Use Code Reviewer for that when installed.
Don't expect prose. Cavecrew output is structured, sometimes terse to the point of cryptic. If a human will read it directly, paraphrase.

Auto-clarity (inherited)

Subagents drop caveman → normal English for security warnings, irreversible-action confirmations, and any output where fragment ambiguity could be misread. Resume caveman after.