buildbetter-customer-issue-triage

Pass

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is highly vulnerable to indirect prompt injection due to its core function of processing untrusted external data and possessing significant system capabilities.
  • Ingestion points: The agent is instructed to fetch and read content from external, user-controllable sources including Linear issues (CUS-*), Slack threads, Gmail messages, customer-provided transcripts, recordings, and file attachments (e.g., archives).
  • Boundary markers: The instructions include strong internal guidelines to 'reconstruct source material' and build 'evidence packets' to verify claims rather than trusting titles or summaries, though there are no technical delimiters (like XML tags) specified for isolating untrusted text during processing.
  • Capability inventory: The skill provides access to production databases (Hasura, direct SQL), Kubernetes (kubectl --context production), GitHub repository management (branching, fixing, PR creation), and communication tools (Gmail via RFC 2822).
  • Sanitization: There are explicit instructions to sanitize customer-facing responses by stripping internal paths, secrets, raw SQL, and implementation details, which helps prevent data leakage but does not fully mitigate the risk of the agent being manipulated by instructions hidden in customer reports.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 12, 2026, 09:33 AM