the-fit-scorer

Pass

Audited by Gen Agent Trust Hub on Aug 14, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill proactively addresses prompt injection risks by incorporating a comprehensive agent-security.md reference. This file mandates that the agent treat all external data as content to be described, never as instructions to be followed, and explicitly forbids following URLs or instructions embedded in fetched content.\n- [INDIRECT_PROMPT_INJECTION]: The skill demonstrates a strong security posture against indirect prompt injection by documenting clear handling procedures for untrusted data.\n
  • Ingestion points: The skill ingests user-provided account lists and enriched firmographic/behavioral signals.\n
  • Boundary markers: references/agent-security.md establishes clear boundaries, instructing the agent to treat retrieved bytes as untrusted data and report embedded instructions as findings.\n
  • Capability inventory: The skill is limited to information processing and table generation; it does not request high-risk capabilities like file system writes (outside .agents/), network requests to arbitrary domains, or subprocess execution.\n
  • Sanitization: The skill specifies HTML escaping for interpolated values and URL validation to ensure that any data passed into markup is safe from XSS.\n- [CREDENTIALS_UNSAFE]: The skill provides specific rules to protect against credential exposure, instructing the agent to identify potential API keys or tokens in CRM exports and notify the user to remove and rotate them, rather than reproducing or persisting them.\n- [DATA_EXPOSURE_AND_EXFILTRATION]: Strict compliance rules are embedded to prevent the use of special category data (e.g., health, race, religion) in scoring models, including a prohibition on using proxies for such data.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 14, 2026, 12:03 PM