Caveman

Use a concise response style while retaining technical facts, identifiers and uncertainty.

Choose a focused investigator, builder or reviewer subagent for a coding task.

Compress the prose in a specified memory file while preserving code and a readable backup.

Draft concise Conventional Commit messages focused on intent.

Identify and label application LLM workflows for Caveman Cloud cost attribution.

Review current Caveman Cloud traces, costs, latency and optimization evidence.

Locate relevant repository code and return compact file-and-line citations.

Review measured token-cost findings and propose consent-gated changes.

Inspect Caveman Cloud experiment states and manage their approval and lifecycle.

Turn a Caveman optimization observation into an evaluated candidate.

Write concise code-review findings with location, problem and proposed fix.

Connect an application’s LLM calls to the Caveman Cloud gateway for measurement.

Show recorded session token usage where the host supports it.

Investigate ambiguous failures with evidence-ranked hypotheses before editing.

Plan and implement a narrow feature by tracing ownership and acceptance criteria.

Plan compatibility-preserving migrations with staged rollout and rollback.

Restructure code while preserving the defined behavior boundary.

Trace a bug to the responsible layer and make a focused change.

Check acceptance conditions with focused evidence and report unresolved risks.