JuliusBrussee/caveman

Cavecrew provides a decision guide for when to spawn caveman-style subagents — investigator, builder, reviewer — to locate code, make small edits, or review diffs, returning compressed output that saves main context. Use it when delegating tasks in Claude Code to reduce token usage.

Caveman compresses AI output into terse, technically accurate caveman-style responses, cutting token usage by 65% while preserving meaning. Use when the user requests caveman mode, brevity, or token efficiency.

Caveman Commit generates concise Conventional Commits messages that cut noise while preserving intent and reasoning. Use it when asked to write a commit message or when staging changes.

Caveman Compress compresses natural language memory files (CLAUDE.md, todos, preferences) into caveman-speak to save input tokens, preserving code, URLs, paths, and structure. It backs up the original and overwrites the file.

Caveman Discover finds every LLM workflow in the current repository and labels it so Caveman Cloud groups spend by what the code actually does. Use when the user asks to discover workflows or break LLM spend down by workflow.

Caveman Evidence Review reads Caveman Cloud data read-only, covering costs, Cave Score, Cave Plan, workflows, traces, latency, errors, compression, routing, and verified savings. Use it when you need to know where LLM spend goes, why cost or quality changed, or which workflows need attention.

Caveman Experiment Manager inspects Caveman Cloud's eval-gated experiment lifecycle, reads current state and results, and reports one supported recommendation or blocks unsafe actions. Use it when asked to start, approve, cancel, promote, or roll back a Caveman experiment.

Caveman Explore is a read-only repository explorer that returns compact path:line citations for where code lives. Use it proactively for cold-start exploration, broad cross-file localization, or when a direct search has failed and you need to find where something lives.

Caveman Help displays a quick-reference card for all caveman modes, skills, and commands. Use it to recall trigger phrases and configuration options without changing the active mode.

Caveman Learn reviews ranked token sinks from a caveman learn report and applies cost-lowering fixes such as trimming heavy CLAUDE.md files or offloading recurring context to cavemem, with per-edit consent. Use it after running 'caveman learn' or when asked to lower an agent's token cost, trim a heavy CLAUDE.md, or offload context that is re-pasted every session.

Caveman Optimize turns Caveman's exact report-only repository observations into an operator-chosen optimization candidate with a paired baseline/candidate evaluation. Use when a Claude Code session needs to inspect an optimization observation, evaluate a candidate change, or act on the current Caveman optimization report.

Caveman Review generates ultra-compressed code review comments that cut noise from PR feedback while preserving actionable signal. Use it when reviewing pull requests or diffs.

Caveman Setup wires the current repository through the Caveman Cloud gateway so every LLM request is measured — cost, tokens, latency — with zero behavior change. Use when the user wants LLM spend observability added to an app or says 'set up caveman'.

Caveman Stats shows real token usage and estimated savings for the current Claude Code session by reading the session log directly. Use /caveman-stats to see net savings after rule overhead, without AI estimation.

Investigate First diagnoses ambiguous failures before editing code. Use it when the cause is unknown, behavior is intermittent, or you need evidence-ranked hypotheses before changing product code.

Lean Build guides feature work with high overbuilding risk, enforcing repository reuse, strict scope, and an explicit stop condition. Use it for new behavior, product slices, or integrations where a minimal, complete path matters.

Migration implements reversible compatibility-safe transitions for schema, data, API, protocol, configuration, or dependency changes. Use it when rollback and preservation proof are required.

Safe Refactor restructures code while preserving behavior. Use it for extraction, consolidation, ownership moves, or cleanup where verification must bracket structural edits.

Surgical Patch fixes bugs and small behavior changes at the narrowest responsible layer. Use Surgical Patch when regression proof, preserved surrounding behavior, and task-relevant tests matter.

Verify and Stop proves existing work meets acceptance conditions without expanding scope. Use it for validation-only tasks, completion checks, focused gate runs, and last-mile proof.