dl-knowledge
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes repository content including source code, test evidence, and draft documents to update the OpenWiki and Architecture Decision Records (ADR). This creates a surface where malicious instructions embedded in these repository files could influence the agent's behavior during the archiving process.
- Ingestion points: The skill reads verified code, acceptance evidence, and draft lists as specified in
SKILL.mdandreferences/knowledge.md. - Boundary markers: While the instructions emphasize verification against the current source code, they do not establish strict delimiters or instructions to ignore embedded prompts within the data being processed.
- Capability inventory: The agent has permissions to read repository files and use specialized tools to update documentation (e.g.,
submit_page,finish). - Sanitization: The skill relies on consistency checks with the source code but lacks explicit sanitization or filtering logic for the text being extracted from repository drafts before it is interpolated into prompts.
Audit Metadata