Reduce Coding-Agent Output Cost With Honey
Warn
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill directs users to install software through agent-specific plugin commands and marketplace additions targeting the Green-PT/honey-for-devs repository, which is not an identified trusted source.
- [METADATA_POISONING]: The skill's frontmatter contains a 'verification' field claiming to be 'security_reviewed', a self-reported status that could lead to misjudgment of the skill's safety.
- [INDIRECT_PROMPT_INJECTION]: The skill uses external documentation to define agent output behavior, representing an indirect injection surface.
- Ingestion points: Processes instructions from AGENTS.md and INSTALL.md at the project root.
- Boundary markers: No explicit delimiters are present to isolate these external instructions from core logic.
- Capability inventory: Modifies the density and stylistic constraints of the agent's code and prose output.
- Sanitization: The skill lacks validation or sanitization mechanisms for the ingested instructional content.
Audit Metadata