debug-fe-be-integration

Pass

Audited by Gen Agent Trust Hub on Aug 25, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill instructions include reading from terminals/*.txt to analyze backend logs. This access pattern exposes potentially sensitive information such as authorization tokens, API keys, and internal system details often found in application logs. It also uses the supabase:execute_sql tool to interact with production databases, which provides a high-privilege path to access or modify sensitive data.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external error reports and terminal logs. This content is used to guide the agent's logic, creating a risk that malicious data within the logs could attempt to override the agent's instructions.
  • Ingestion points: Terminal logs (terminals/*.txt) and Sentry API issue search results.
  • Boundary markers: None. The instructions do not specify any delimiters to isolate untrusted data from the system prompt.
  • Capability inventory: The agent can perform file system reads, web searches via firecrawl, and execute SQL queries via the Supabase tool.
  • Sanitization: The skill does not include steps to sanitize or filter the content of the logs before processing.
  • [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill uses external tools like firecrawl:firecrawl_search to fetch information from the web. While used for researching debugging patterns, this involves outgoing network requests to external services.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 25, 2026, 05:56 AM
Security Audit — agent-trust-hub — debug-fe-be-integration