talk-stage4-position
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external files which could potentially contain malicious instructions intended to subvert the agent's logic. While the impact is limited by the skill's restricted toolset and the requirement for user confirmation, the ingestion process lacks protective delimiters.\n
- Ingestion points: The skill reads
talks/{YYYY}-{slug}-summary.mdandtalks/{YYYY}-{slug}-concepts.mdusing theReadtool.\n - Boundary markers: Absent. The instructions do not define markers to separate the ingested talk data from the agent's core instructions.\n
- Capability inventory: The skill employs
Writefor file creation andAskUserQuestionfor user interaction. It lacks dangerous capabilities like shell execution or network access.\n - Sanitization: Absent. There is no specified logic for validating or escaping the contents of the files before they are processed by the LLM.
Audit Metadata