Update to Caveman

{
  "name": "cavecrew",
  "description": "Use when a coding task should be delegated to a focused investigator, small-change builder, or diff reviewer so the main context stays compact.",
  "included_files": [
    {
      "relative_path": "README.md",
      "size_in_bytes": 2873
    },
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 226
    }
  ],
  "skill_md_contents": "---\nname: cavecrew\ndescription: Use when a coding task should be delegated to a focused investigator, small-change builder, or diff reviewer\n  so the main context stays compact.\n---\n\nCavecrew = three subagent presets that emit caveman output. Same job as Anthropic defaults (`Explore`, edit-style agents, reviewer); difference is the tool-result they return is compressed, so main context shrinks per delegation.\n\n## When to use cavecrew vs alternatives\n\n| Task | Use |\n|---|---|\n| \"Where is X defined / what calls Y / list uses of Z\" | `cavecrew-investigator` |\n| Same but you also want suggestions/architecture commentary | `Explore` (vanilla) |\n| Surgical edit, ≤2 files, scope obvious | `cavecrew-builder` |\n| New feature / 3+ files / cross-cutting refactor | Main thread or `feature-dev:code-architect` |\n| Review diff, branch, or file for bugs | `cavecrew-reviewer` |\n| Deep code review with rationale + alternatives | `Code Reviewer` (vanilla) |\n| One-line answer you already know | Main thread, no subagent |\n\nRule of thumb: **if you'd want the subagent's output in 1/3 the tokens, pick cavecrew. If you'd want prose, pick vanilla.**\n\n## Why this exists (the real win)\n\nSubagent tool results get injected into main context verbatim. A vanilla `Explore` that returns 2k tokens of prose costs 2k tokens of main-context budget every time. The same finding from `cavecrew-investigator` returns ~700 tokens. Across 20 delegations in one session that's the difference between context exhaustion and finishing the task.\n\n## Output contracts\n\nWhat main thread can rely on per agent:\n\n**`cavecrew-investigator`**\n```\n<Header>:\n- path:line — `symbol` — short note\ntotals: <counts>.\n```\nOr `No match.` Always file-path-first, line-number-attached, backticked symbols. Safe to grep with `path:\\d+`.\n\n**`cavecrew-builder`**\n```\n<path:line-range> — <change ≤10 words>.\nverified: <re-read OK | mismatch @ path:line>.\n```\nOr one of: `too-big.` / `needs-confirm.` / `ambiguous.` / `regressed.` (terminal first token).\n\n**`cavecrew-reviewer`**\n```\npath:line: <emoji> <severity>: <problem>. <fix>.\ntotals: N🔴 N🟡 N🔵 N❓\n```\nOr `No issues.` Findings sorted file → line ascending.\n\n## Chaining patterns\n\n**Locate → fix → verify** (most common):\n1. `cavecrew-investigator` returns site list.\n2. Main thread picks 1-2 sites, hands paths to `cavecrew-builder`.\n3. `cavecrew-reviewer` audits the diff.\n\n**Parallel scout** (when investigation is broad):\nSpawn 2-3 `cavecrew-investigator` calls in one message (different angles: defs vs callers vs tests). Aggregate in main thread.\n\n**Single-shot edit** (when site is already known):\nSkip investigator. Hand exact path:line to `cavecrew-builder` directly.\n\n## What NOT to do\n\n- Don't use `cavecrew-builder` when you don't already know the file. Spawn investigator first or main thread will eat tokens passing context.\n- Don't chain `cavecrew-investigator → cavecrew-builder` for a 5-file refactor. Builder will return `too-big.` and you'll have wasted a turn.\n- Don't ask `cavecrew-reviewer` for \"general feedback\" — it returns findings only, no architecture opinions. Use `Code Reviewer` for that.\n- Don't expect prose. Cavecrew output is structured, sometimes terse to the point of cryptic. If a human will read it directly, paraphrase.\n\n## Auto-clarity (inherited)\n\nSubagents drop caveman → normal English for security warnings, irreversible-action confirmations, and any output where fragment ambiguity could be misread. Resume caveman after.\n"
}