distribute

Use this skill for $brew distribute <skill-path>.

Objective

Evaluate a CocoPod skill package before marketplace distribution or recommendation. The gate protects users from prompt-dump skills, undeclared permissions, install-time execution, untested behavior, and unjustified network use.

Five Criteria

1. Explicit tool permissions: SKILL.md frontmatter must declare the tools or permissions the skill expects.
2. Observable evals: evals/ must exist and contain at least one runnable evaluation.
3. Failure behavior: SKILL.md must document on_failure: behavior or an equivalent failure-handling section.
4. No install-time execution: The package must not execute code during installation. Execution belongs only to explicit command invocations.
5. Justified network use: Any network: frontmatter declaration must include network_justification:.

Verdicts

• Clear: All quality criteria pass and no security gate fails. Distribution may proceed.
• Advisory: One or more quality criteria are incomplete, but no security gate fails. Distribution requires developer acknowledgment and is not recommended automatically by CocoBrew.
• Block: A security criterion fails, install-time execution is detected, or the package omits critical permission declarations. Distribution stops. There is no bypass flag.

Workflow

1. Parse $brew distribute <skill-path>.
2. Read SKILL.md and package folders next to it.
3. Run deterministic checks before any LLM review:
   • frontmatter contains permissions/tools
   • evals/ exists and is non-empty
   • failure handling section exists
   • no install hooks or install-time command blocks are declared
   • network declarations include justification
4. Run existing trifecta/security checks when available.
5. Write .cocoplus/brew/distribution-gate-<timestamp>.json.
6. For Clear, proceed with the configured distribution path.
7. For Advisory, stop for acknowledgment.
8. For Block, stop with exact failed criteria.

Exit Criteria

• Block verdict prevents distribution.
• Advisory verdict requires explicit acknowledgment.
• The report names each failed criterion.
• No install-time execution is allowed.

Anti-Rationalization

Shortcut / Temptation	Why It Fails
Treat the skill as complete because the file exists	Skill contracts must describe observable behavior and verification, not just command names.
Skip artifact and safety checks for a small command	Small commands still mutate state or guide execution; preserve the same gates.