corpus
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill manages a workflow for processing untrusted external data, which creates an indirect prompt injection vulnerability surface.\n
- Ingestion points: The skill ingests external files and directories into the
corpus/directory as part of thereferences/ingest.mdworkflow.\n - Boundary markers: Both
references/consult.mdandreferences/ingest.mdinclude explicit instructions for the agent to treat source content as "untrusted data" and to specifically avoid following instructions found within evidence.\n - Capability inventory: The skill is capable of reading local files, writing to the
corpus/directory, modifying the repository's.gitignorefile, and updating or creating the agent instruction fileAGENTS.md.\n - Sanitization: Ingestion workflows in
references/ingest.mdmandate faithful extraction of text without added interpretation or summaries, which helps mitigate the risk of the agent being misled by malicious content during the ingestion process.
Audit Metadata