Guides delegation to compressed-output subagents for code location, editing, and review tasks.
npx skills add https://github.com/TrayMachi/dotfiles --skill cavecrew-traymachi
Please help me install this Agent Skill. Skill: cavecrew Source: https://github.com/TrayMachi/dotfiles/tree/main/agents/skills/cavecrew Command: npx skills add https://github.com/TrayMachi/dotfiles --skill cavecrew-traymachi
cavecrew-investigator for locating code, cavecrew-builder for surgical 1-2 file edits, cavecrew-reviewer for diff audits, or vanilla agents when prose and rationale are wanted. - Output Contracts: Defines exact structured formats each subagent returns (path:line findings, verified edit confirmations, severity-tagged review findings) so the main thread can parse results reliably. - Chaining Patterns: Documents workflows like locate → fix → verify and parallel scouting with multiple investigators. - Use Case: During a long refactoring session, 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 keeping main-context usage minimal. ## Quick Start Ask the agent to delegate a code search to the cavecrew investigator instead of running the exploration inline to save context.High-intent search queries and answers about installing and using this skill.
Spawn cavecrew-investigator with a locate-style query such as finding a definition or listing callers. It returns compressed path:line findings with backticked symbols, roughly one third the size of vanilla Explore output, preserving main-context budget.
Use cavecrew-builder for surgical edits of one to two files where the scope is already obvious. For three or more files, cross-cutting refactors, or new features, use the main thread or a feature architect agent, since the builder refuses oversized scope with a too-big response.
cavecrew-reviewer returns one-line findings with severity emoji sorted by file and line, with no architecture commentary. The vanilla Code Reviewer provides prose, rationale, and alternatives, which costs more context but suits deep reviews.
Yes, the parallel scout pattern spawns two to three cavecrew-investigator calls in one message with different angles such as definitions, callers, and tests. The main thread then aggregates the compressed results.
Avoid cavecrew when you want prose, architecture opinions, or rationale, when the answer is a one-liner you already know, or when a human will read the output directly. The terse structured format can be cryptic without paraphrasing.
rvdbreemenRoute code tasks to investigator, builder, or reviewer subagents.
DieStokSimplify changed source code while preserving behavior and passing tests.
SimonZimmerGenerates one-line code review comments with location, severity, problem, and fix.
mmdmcyRoute large payloads to subagents and keep summaries in the main context window.