appointment-setting
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow that ingests untrusted data from external websites and user-provided conversation history without adequate isolation.
- Ingestion points:
SKILL.mdinstructs the agent to "fetch the site or page they named" and to process "The conversation so far" provided by the user. - Boundary markers: The instructions do not specify the use of delimiters, XML tags, or other markers to separate this external data from the skill's core logic.
- Capability inventory: The agent is authorized to fetch web content and perform file I/O operations (read and write) on the
.agents/product-context.mdfile. - Sanitization: There are no explicit requirements to sanitize, filter, or validate the ingested content before it is used for writing copy or saved to persistent storage.
- [INDIRECT_PROMPT_INJECTION]: The skill uses a persistent context file that could be susceptible to gradual poisoning.
- Evidence:
SKILL.mdinstructs the agent to "Write what you learn [from site research] to .agents/product-context.md so the next skill does not repeat the work." If a malicious website is researched, injected instructions could be stored in this file and executed by other skills that reference the same context.
Audit Metadata