competitive-intelligence-gathering

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The extract_intel_with_llm function in SKILL.md is vulnerable to indirect prompt injection. It interpolates the conversation object, which contains untrusted prospect input, directly into a prompt template without sanitization or clear boundary markers. This could allow an adversary to influence the agent's extraction logic or subsequent actions if the injected instructions are followed by the model.\n
  • Ingestion points: The conversation parameter in the extract_intel_with_llm function (SKILL.md).\n
  • Boundary markers: Absent. The prompt uses a simple f-string interpolation: {format_conversation(conversation)} without delimiters like XML tags or triple backticks to isolate the untrusted data.\n
  • Capability inventory: The skill includes functions for database storage (store_intel), alerting (trigger_alert_if_needed), and updating battlecards (update_battlecard_from_intel). While these are internal functions, a successful injection could manipulate the competitive data stored in the organization's database.\n
  • Sanitization: Absent. No sanitization, validation, or filtering of the conversation content is performed before interpolation into the prompt.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 04:04 PM
Security Audit — agent-trust-hub — competitive-intelligence-gathering