at-most-once-publishing
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill outlines patterns for processing external data for publication, including generating identity hashes from 'canonical request projections' and 'normalized visible bodies'. This describes a potential attack surface where untrusted external content enters the agent's workflow.
- Ingestion points: Processes described in
SKILL.mdfor deriving versions and identities from external source content (canonical URLs or object IDs). - Boundary markers: No explicit instructions for implementing prompt delimiters or 'ignore' instructions for the content being published.
- Capability inventory: The skill is configured with
Bash,Read,Write,Edit,Grep, andGlobtools. - Sanitization: The skill advocates for normalization of content before hashing, which serves as a form of structural validation but not necessarily prompt-injection sanitization.
Audit Metadata