skills.rest
leogurja
leogurjaCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

cavecrew

Decide when to delegate code tasks to compressed-output subagents instead of working inline.

1|Updated May 3, 2021
Search Tags:#code review#claude code#delegation#code search#subagent#context optimization#diff review
One-click install
npx skills add https://github.com/leogurja/dotfiles --skill cavecrew-leogurja
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: cavecrew
Source: https://github.com/leogurja/dotfiles/tree/main/home/dot_agents/skills/cavecrew
Command: npx skills add https://github.com/leogurja/dotfiles --skill cavecrew-leogurja

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Subagent results are injected verbatim into the main conversation context, so verbose agent output quickly exhausts the context window during long coding sessions. This Skill provides a decision guide for when to delegate to three compressed-output subagents (investigator, builder, reviewer) versus working inline or using vanilla agents, keeping main-context usage low across many delegations. ## Core Features & Use Cases - Delegation decision matrix: Maps task types (locate code, surgical edit, diff review) to the right subagent or to vanilla agents like Explore and Code Reviewer when prose is preferred. - Three subagent presets: cavecrew-investigator returns path:line site lists, cavecrew-builder performs 1-2 file edits with verified results, and cavecrew-reviewer emits severity-tagged findings with totals. - Chaining patterns: Supports locate-fix-verify pipelines, parallel scouting with multiple investigators, and single-shot edits when the site is already known. - Use Case: During a large refactor, spawn cavecrew-investigator to find all call sites of a function, hand the top two paths to cavecrew-builder for the edit, then have cavecrew-reviewer audit the diff, all while consuming a fraction of the context a vanilla agent would. ## Quick Start Ask the assistant to delegate code location, a small edit, or a diff review to a cavecrew subagent instead of doing it inline.

Frequently Asked Questions about cavecrew

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
When should I delegate to a subagent instead of working inline?▼

Delegate when the subagent's output can be compressed, such as locating code definitions, making a 1-2 file edit, or reviewing a diff. Work inline for one-line answers you already know and for 3+ file refactors that exceed builder scope.

How do I chain investigator, builder, and reviewer subagents?▼

Run cavecrew-investigator to get a path:line site list, pick 1-2 sites and hand them to cavecrew-builder for the edit, then have cavecrew-reviewer audit the resulting diff. This locate-fix-verify chain is the most common pattern.

What is the difference between cavecrew-investigator and the Explore agent?▼

cavecrew-investigator returns a compressed path:line list with symbols and short notes, while Explore returns prose with architecture commentary and suggestions. Choose investigator to save context tokens and Explore when you want rationale.

Can I change the model used by cavecrew subagents?▼

Yes, set CAVECREW_REVIEWER_MODEL, CAVECREW_BUILDER_MODEL, or CAVECREW_INVESTIGATOR_MODEL environment variables before launching Claude Code. The override patches only the model line in the installed agent frontmatter.

What are the limitations of the cavecrew-builder subagent?▼

cavecrew-builder refuses scopes of 3 or more files and returns terminal signals like too-big, needs-confirm, ambiguous, or regressed. It requires that you already know the target file, so spawn an investigator first if the site is unknown.

Related Skills

Explore More Software Engineering Skills→
Space-DinosaursSpace-Dinosaurs
12

agent-engineer

Automate end-to-end code-change implementation across repositories with quality gates.

Official
Advanced
AllBeingsFutureAllBeingsFuture
9

allbeingsfuture/skills@code-review

Reviews local git diffs and remote Pull Requests for correctness, security, and maintainability issues.

Community
Intermediate
uzumaki-incuzumaki-inc

health

Run code health checks and compute a weighted 0-10 score.

Official
Advanced
amirkiarafieiamirkiarafiei
5

qwen-code

Delegate large multi-step repository edits to Qwen-Code CLI with shared-state handoff.

Community
Advanced