writing-beats
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted 'raw material' from a user-supplied markdown file to generate content. This creates a vulnerability surface where malicious instructions could be embedded in the raw data. \n
- Ingestion points: The markdown file containing raw material provided by the user, as described in
SKILL.md. \n - Boundary markers: Absent. There are no instructions for the agent to use delimiters or to treat the raw material strictly as data. \n
- Capability inventory: The agent has the ability to read from the file system and write/append to an article file, as outlined in the writing loop in
SKILL.md. \n - Sanitization: Absent. No sanitization or validation logic is defined for the ingested material.\n- [METADATA_POISONING]: The skill uses the term 'exploit' in its description and instructions. Although used in a metaphorical sense related to the 'explore-exploit' tradeoff in data processing, such terminology is typically associated with malicious activities and could be used to camouflage intent or trigger false positives in security systems.
Audit Metadata