call-preparation

Pass

Audited by Gen Agent Trust Hub on Aug 27, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external sources, creating a potential surface for indirect prompt injection attacks.
  • Ingestion points: The agent is instructed to fetch and read external websites for company research (SKILL.md, line 14) and ingest user-pasted meeting transcripts or notes (SKILL.md, line 67).
  • Boundary markers: The skill explicitly references references/agent-security.md, which contains a dedicated 'Rule 1' stating that fetched/pasted content must be treated as data and never as instructions. It mandates quoting suspected injection attempts rather than executing them.
  • Capability inventory: The skill has the capability to read from and write to the .agents/product-context.md file, fetch external URLs, and process text inputs.
  • Sanitization: The referenced references/agent-security.md (Rule 3) requires HTML-escaping of all interpolated values from untrusted sources before they are emitted in markup or code.
  • [PROMPT_INJECTION]: The skill contains instructions that direct the agent to conceal specific types of information from the user, which can be misused to hide safety or quality caveats.
  • Evidence: SKILL.md (lines 78-79) states: "The user copies the draft and leaves the warning behind, so a caveat protects you and not them. Do not return a draft with a warning on it." and "Run the input list below before you write anything. If one of those inputs is missing, ask for it and stop."
  • Context: While primarily aimed at ensuring high-quality, complete outputs for sales professionals, these instructions explicitly forbid the inclusion of warnings in drafts, which could result in the omission of important contextual or safety-related disclosures if the model proceeds with assumptions.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 27, 2026, 08:37 AM