letterly-export

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill establishes an ingestion surface by downloading data from an external web application and saving it directly to the local filesystem without sanitization. This could serve as a vector for indirect prompt injection if the downloaded content is subsequently processed by an AI agent.
  • Ingestion points: CSV data downloaded from web.letterly.app in scripts/export.py.
  • Boundary markers: None; the skill saves the raw external data to the vault without adding delimiters or warning markers.
  • Capability inventory: The skill possesses browser automation capabilities via Playwright and file system write access.
  • Sanitization: The script does not perform validation or filtering on the contents of the downloaded CSV files.
  • [DATA_EXFILTRATION]: While the skill primarily moves data from the web to a local vault, it accesses a persistent browser context containing sensitive user information such as cookies and session tokens.
  • Evidence: scripts/export.py initializes p.chromium.launch_persistent_context using a user data directory provided by a shared utility, granting the script access to the user's browser profile.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 10:16 PM
Security Audit — agent-trust-hub — letterly-export