signals-scout-surveys

Pass

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted open-text feedback from survey respondents, which presents a surface for potential indirect prompt injection.
  • Ingestion points: Survey responses are ingested via execute-sql calls to the events table as outlined in SKILL.md and references/response-querying.md.
  • Boundary markers: The skill lacks explicit data delimiters but enforces logic-based boundaries by requiring convergence from multiple distinct respondents (at least 5, or 3 if correlated with activity logs) before a theme is considered report-worthy.
  • Capability inventory: The agent is authorized to use emit_report and edit_report to send data to the project inbox, and scout-scratchpad-remember to persist data between runs.
  • Sanitization: The instructions explicitly require the agent to strip personally identifiable information (PII) from responses and truncate verbatim quotes to the sentence level.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 10, 2026, 04:00 PM
Security Audit — agent-trust-hub — signals-scout-surveys