cavecrew

When to delegate to `cavecrew-investigator` (locate code), `cavecrew-builder` (1-2 file edit) or `cavecrew-reviewer` (diff review) instead of working inline or using `Explore`. Their output is compressed, so main context lasts longer.

Quality

82%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Quality

Content

86%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

An unusually disciplined body: exact output contracts, a routing table with named alternatives, three concrete chaining patterns, and an explicit anti-pattern list, all with essentially no filler. The only deductions are mild repetition of the compression rationale across three places and the absence of a recovery step after the reviewer's audit.

Dimension	Reasoning	Score
Conciseness	The body is dense and assumes competence ("Same job as Anthropic defaults… difference is the tool-result they return is compressed"), but the compression/context tradeoff is restated across the intro paragraph, the "Rule of thumb" line, and "Why this exists" — minor instances that could be trimmed. Fits anchor 4 rather than 5 (some repetition) and clearly not 3 (no padding or explanations of known concepts).	4 / 5
Actionability	Provides fully concrete, executable guidance: exact output-contract templates ("<path:line-range> — <change ≤10 words>. verified: <re-read OK \| mismatch @ path:line>"), enumerated terminal tokens ("too-big." / "needs-confirm." / "ambiguous." / "regressed."), sort orders, and spawn counts ("Spawn 2-3 `cavecrew-investigator` calls in one message"). This matches anchor 5; per the code-vs-instruction note, absence of code is not penalized when guidance is this precise.	5 / 5
Workflow Clarity	"Locate → fix → verify" is a clearly numbered 3-step sequence with an explicit validation checkpoint ("cavecrew-reviewer audits the diff"; builder's "verified: re-read OK \| mismatch" re-read check) and anticipated failure tokens. Fits anchor 4 rather than 5 because there is no error-recovery loop after the reviewer flags issues; no destructive/batch cap applies.	4 / 5
Progressive Disclosure	The skill is a self-contained routing guide with no bundle files (references/, scripts/, assets/ are absent) and no external file references; all content (routing table, contracts, chaining patterns, anti-patterns) belongs inline at this size. Sections are well-organized and flat, matching the simple-skill exception for anchor 5 with no organization gaps to justify 4.	5 / 5
	Total	18 / 20 Passed

Description

78%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong, tight description that names all three delegation targets with concrete task scopes, states the core differentiator (compressed output extending main context), and explicitly contrasts itself with inline work and `Explore`. The only gaps are a few missing natural trigger synonyms and slightly abstract "when" phrasing.

Dimension	Reasoning	Score
Specificity	Names three concrete per-agent capabilities — "locate code", "1-2 file edit", "diff review" — plus the differentiator "output is compressed, so main context lasts longer". Matches anchor 4 (several specific actions, minor gaps) rather than 5, since scope limits and edge cases are not covered.	4 / 5
Completeness	Explicitly answers both what ("delegate to `cavecrew-investigator` (locate code), `cavecrew-builder` (1-2 file edit) or `cavecrew-reviewer` (diff review)"; "output is compressed") and when ("When to delegate to … instead of working inline or using `Explore`"). The "when" is explicit routing guidance, so it clears the anchor-3 cap, but it is a routing rule with parenthetical task labels rather than fully concrete trigger phrases, fitting anchor 4 better than 5.	4 / 5
Trigger Term Quality	Includes natural terms a user or model would say — "locate code", "edit", "diff review", "delegate", and the competing preset name "Explore" — but misses common synonyms such as "find where X is defined", "search the codebase", or "review changes". Good coverage with a few natural terms missing fits anchor 4, not 3 (coverage is solid, not partial) and not 5 (no synonyms/extensions).	4 / 5
Distinctiveness Conflict Risk	The description explicitly names the alternatives it competes with ("instead of working inline or using `Explore`") and scopes each agent to a distinct task, giving a clear niche with minimal conflict risk — anchor 5, not 4, since no meaningful overlap with other skills remains.	5 / 5
	Total	17 / 20 Passed

Validation

100%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: JuliusBrussee/caveman
Path: skills/cavecrew/SKILL.md
Commit: f5d7294

Reviewed: 1 hour ago

Table of Contents

Discovery Implementation Validation

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.