objection-handling
Pass
Audited by Gen Agent Trust Hub on Aug 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill contains a capability to ingest data from external websites to build its product context file, which creates a potential attack surface. This is effectively mitigated by the skill's structure and instructions.
- Ingestion points: SKILL.md instructs the agent to fetch websites or pages to build the product context.
- Boundary markers: references/house-rules.md (Rule 7) provides explicit instructions to treat all fetched content as data and never as instructions.
- Capability inventory: The skill reads from the web and writes extracted data to the .agents/product-context.md local file.
- Sanitization: Rule 7 in house-rules.md requires quoting any instructions found within the data to prevent execution.
- [EXTERNAL_DOWNLOADS]: The skill fetches content from external URLs provided by the user to perform market and competitor research. This process is governed by specific pack-wide house rules designed to prevent the execution of malicious instructions embedded in fetched pages.
- [PROMPT_INJECTION]: A deterministic detector flagged a concealment pattern regarding the instruction 'Do not return a draft with a warning on it'. However, analysis confirms this is a benign quality requirement. It ensures that output intended for external use (copywriting) does not contain meta-commentary or warnings that the user might inadvertently include when copying the content for a prospect.
Audit Metadata