Route code tasks to investigator, builder, or reviewer subagents.
npx skills add https://github.com/rvdbreemen/OTGW-firmware --skill cavecrew-rvdbreemen
Please help me install this Agent Skill. Skill: cavecrew Source: https://github.com/rvdbreemen/OTGW-firmware/tree/main/.agents/skills/cavecrew Command: npx skills add https://github.com/rvdbreemen/OTGW-firmware --skill cavecrew-rvdbreemen
It prevents context exhaustion during long coding sessions by helping you delegate investigation, small edits, and diff reviews to caveman-style subagents that return compressed tool results.
cavecrew-investigator, cavecrew-builder, and cavecrew-reviewer based on task size, uncertainty, and desired output format.Example: When you ask “Where is the MQTT MsgID routing logic defined and what calls it?”, use cavecrew-investigator to get a grep-friendly file:line list, then hand 1–2 exact paths to cavecrew-builder for a targeted fix, and finally run cavecrew-reviewer to audit the resulting diff.
Tell the AI: “delegate to cavecrew for a compressed investigation of where X is defined, then review the diff for issues.”
High-intent search queries and answers about installing and using this skill.
You can delegate code investigation and editing tasks to subagents that return compressed tool outputs, preventing context exhaustion. This preserves your main thread tokens by routing work to specialized agents that summarize findings and diffs efficiently.
Use a delegation decision guide to route tasks to specialized subagents based on size and uncertainty. Assign investigators for locating definitions, builders for surgical edits, and reviewers for structured diff findings with severity levels.
Delegate the task to a builder subagent after providing exact file paths identified during investigation. The builder executes targeted changes while respecting the explicit file-scope limit, ensuring edits remain constrained to one or two files.
Yes, you can delegate the review process to a reviewer subagent. It audits the resulting code diff and produces a structured report highlighting issues and their respective severity levels, ensuring changes are thoroughly evaluated.
Use subagent delegation for code navigation tasks when you need to locate definitions or callers without exhausting your main context. Delegating to an investigator subagent returns a compressed file and line list, saving tokens for subsequent work.
No, cavecrew does not require external dependencies to manage agent orchestration. It operates as a standalone delegation decision guide that triggers on specific phrases to route tasks to investigation, builder, or reviewer subagents using compressed output contracts.
samjhechtGenerate task-level implementation plans with exact file paths and code samples.
reedmayhew18Identify issues in code changes and produce a severity-organized report.
Yahia89Create persistent, stateful AI agents with thread management and streaming in Convex.
achmfDelegate coding tasks to specialized subagents that return compressed outputs.