journal-of-the-acm
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest data from external local files (
../../resources/source-basis.mdand../../resources/official-source-map.md) to guide its advice. This establishes a data ingestion surface where external content is interpolated into the agent's context. - Ingestion points:
SKILL.md(Official-submission checklist section) directs the agent to read specific resource files using relative paths. - Boundary markers: Absent; the instructions do not provide delimiters, XML tags, or 'ignore' directives to prevent the agent from following instructions potentially contained within those external files.
- Capability inventory: The skill utilizes the agent's file-reading capabilities (e.g.,
read_fileor equivalent tools) to access paths outside the skill's own directory. - Sanitization: No sanitization, escaping, or validation logic is defined to check the content read from these external files before processing.
Audit Metadata