hrt-artifact-lint
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill parses content from user-controlled markdown files and instructs the agent to automatically 'fix' structural errors. This creates a potential surface where maliciously crafted content in the files being linted could influence the agent through the tool's error messages. \n
- Ingestion points: The skill reads
proposal.md,design.md,tasks.md, and various files within thespecs/directory via thelint.cjsscript. \n - Boundary markers: The instructions lack explicit boundary markers or warnings to ignore instructions embedded within the data being linted. \n
- Capability inventory: The agent is instructed to modify files on disk ('Fix it directly, no user prompt') based on the tool's error output, which is a write capability. \n
- Sanitization: While the script uses regular expressions to extract structured data, it includes user-defined strings (such as requirement and capability names) directly in its output report without filtering for potential prompt injection patterns.
Audit Metadata