price-negotiation

Pass

Audited by Gen Agent Trust Hub on Aug 27, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCEDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill establishes a process for fetching data from user-specified external websites and reading local context files, creating a surface for indirect prompt injection.
  • Ingestion points: External site content fetched during the research phase and data stored in .agents/product-context.md (SKILL.md).
  • Boundary markers: The skill relies on a mandatory house-rules.md (Rule 7) which explicitly instructs the agent to treat all fetched, pasted, or transcribed content as data and never as instructions, requiring the agent to quote rather than act on embedded commands.
  • Capability inventory: The skill is capable of writing research data to the filesystem (.agents/product-context.md) and chaining execution to other skills such as opportunity-scoring and contribution-margin.
  • Sanitization: The agent is required to perform a nine-question quality and safety check before returning any output to the user.
  • [PERSISTENCE]: The skill utilizes a local file storage mechanism to maintain state across different sessions or skill executions.
  • Mechanism: The instructions in SKILL.md mandate writing researched company information to .agents/product-context.md so that subsequent skills can access the same context without repeating the research work.
  • [DATA_EXFILTRATION]: The skill prompts the user for sensitive business information, creating a potential exposure surface for trade secrets.
  • Sensitive Data: The workflow requests contribution margins, deal sizes, specific negotiation terms, and walk-away points from the user.
  • Mitigation: The house-rules.md (Rule 8) includes a security protocol that requires the agent to flag any credentials detected in inputs for rotation and strictly prohibits their reproduction in the output.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 27, 2026, 08:37 AM