Cavecrewv1.0.0

An engineering team needs to locate all usages of a deprecated API across a 500-file monolith before migrating. They spawn `cavecrew-investigator` in parallel from three angles (definitions, callers, tests) to get compressed site lists, then hand the top paths to `cavecrew-builder` for surgical edits, and finally run `cavecrew-reviewer` on the diff to catch bugs before merge.

A developer is deep in a long debugging session with a 128k-token context window and needs to investigate dozens of trace points without exhausting context. Using cavecrew's compressed tool results (~60% smaller) lets them run 20+ investigator delegations where vanilla `Explore` would have caused context exhaustion halfway through. This keeps the main thread focused on reasoning instead of re-reading verbose subagent prose.

A DevOps team integrates `cavecrew-reviewer` into their pull request pipeline to audit every diff for bugs, severity levels, and suggested fixes in a compressed format. The structured output (`path:line: emoji severity: problem. fix.`) is parsed by a bot that posts inline PR comments, while `No issues.` short-circuits the pipeline to avoid unnecessary human review on trivial changes.

A platform team building an autonomous coding agent uses cavecrew's chaining patterns (locate → fix → verify, parallel scout, single-shot edit) as reference contracts for their own subagent routing. The explicit output contracts and terminal tokens like `too-big.` or `needs-confirm.` provide reliable branching signals for the orchestrator to decide when to escalate to the main thread or a vanilla agent.

A new hire needs to quickly map where key symbols are defined and used in a large codebase. They repeatedly invoke `cavecrew-investigator` for specific questions like 'what calls Y' and get file-path-first, line-number-attached answers that are easy to grep and navigate. For architecture-level questions they know to fall back to vanilla `Explore` to get the prose commentary cavecrew omits.

Cavecrew is distributed freely as an open-source Claude skill to drive adoption among individual developers and small teams. Revenue comes from enterprise support contracts, SLAs, and custom subagent tuning for large engineering orgs that need guaranteed context savings and integration help.

The skill is bundled as a free tier in a broader AI coding platform, with usage limits on total delegations per month. Paid tiers unlock higher delegation quotas, parallel scout limits, custom output contract schemas, and team-wide analytics on context savings.

Cavecrew is licensed to vendors building AI coding assistants (IDE plugins, CLI agents) as an embedded subagent module that reduces inference costs by shrinking tool-result tokens. Vendors pay per active developer seat or per API call routed through cavecrew presets.