cavecrew
Decision guide for delegating to caveman-style subagents. Tells the main thread WHEN to spawn `cavecrew-investigator` (locate code), `cavecrew-builder` (1-2 file edit), or `cavecrew-reviewer` (diff review) instead of doing the work inline or using vanilla `Explore`. Subagent output is caveman-compressed so the tool-result injected back into main context is ~60% smaller — main context lasts longer across long sessions. Trigger: "delegate to subagent", "use cavecrew", "spawn investigator/builder/reviewer", "save context", "compressed agent output".
77
96%
Does it follow best practices?
Impact
—
No eval scenarios have been run
Passed
No known issues
Quality
Discovery
100%Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.
This is a strong skill description that clearly communicates its purpose, specific capabilities, and when to use it. It names concrete subagent types with their roles, explains the compression benefit quantitatively, and provides explicit trigger phrases. The unique terminology (cavecrew, caveman-compressed) makes it highly distinguishable from other skills.
| Dimension | Reasoning | Score |
|---|---|---|
Specificity | Lists multiple specific concrete actions: delegating to named subagents (cavecrew-investigator for locating code, cavecrew-builder for 1-2 file edits, cavecrew-reviewer for diff review), explains the compression benefit (~60% smaller context), and clarifies the decision logic of when to spawn vs. work inline. | 3 / 3 |
Completeness | Clearly answers both 'what' (decision guide for delegating to caveman-style subagents with specific roles) and 'when' (explicit trigger phrases listed, plus contextual guidance about when to spawn vs. work inline or use vanilla Explore). | 3 / 3 |
Trigger Term Quality | Includes natural trigger terms users would say: 'delegate to subagent', 'use cavecrew', 'spawn investigator/builder/reviewer', 'save context', 'compressed agent output'. These cover both the specific tool names and the general intent (saving context, delegating work). | 3 / 3 |
Distinctiveness Conflict Risk | Highly distinctive with unique terminology ('cavecrew', 'caveman-compressed', specific subagent names like cavecrew-investigator/builder/reviewer). Very unlikely to conflict with other skills due to the specialized vocabulary and clear niche of subagent delegation with context compression. | 3 / 3 |
Total | 12 / 12 Passed |
Implementation
92%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This is a high-quality decision-guide skill that efficiently communicates when and how to delegate to cavecrew subagents. Its greatest strengths are the crisp decision table, precise output contracts with parseable formats, and clear chaining patterns with implicit validation. The only notable gap is the lack of references to the actual subagent preset files or configurations, which would strengthen progressive disclosure.
Suggestions
Add references to the actual subagent preset/configuration files (e.g., 'See [cavecrew-investigator.md](cavecrew-investigator.md) for full preset') so the main thread can navigate to detailed setup if needed.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The content is lean and efficient. The 'Why this exists' section borders on explanatory but earns its place by quantifying the token savings (2k vs 700 tokens), which is genuinely useful decision-making context. No concepts are over-explained; the table format and terse output contracts respect Claude's intelligence. | 3 / 3 |
Actionability | The decision table provides concrete, unambiguous guidance for when to use each agent. Output contracts specify exact formats with parseable patterns (e.g., `path:\d+`), terminal tokens for error states (`too-big.`, `needs-confirm.`), and chaining patterns give step-by-step delegation sequences. The 'What NOT to do' section adds critical guardrails with specific failure modes. | 3 / 3 |
Workflow Clarity | The chaining patterns section provides clear multi-step sequences (locate → fix → verify) with explicit validation via the reviewer step. The builder's terminal tokens (`too-big.`, `regressed.`, `mismatch @ path:line`) serve as built-in validation checkpoints with implicit feedback loops. The 'What NOT to do' section acts as an error-prevention checklist. | 3 / 3 |
Progressive Disclosure | The content is well-structured with clear sections and good use of tables and code blocks. However, there are no bundle files and no references to external files for the individual subagent configurations or detailed documentation. The output contracts and chaining patterns could benefit from linking to the actual subagent preset definitions, though the skill is compact enough that this is a minor issue. | 2 / 3 |
Total | 11 / 12 Passed |
Validation
100%Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.
Validation — 11 / 11 Passed
Validation for skill structure
No warnings or errors.
- Repository
- JuliusBrussee/caveman
- Commit
655b7d9
- Reviewed
Table of Contents
Is this your skill?
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.