cursor-transcript-harvest

Pass

Audited by Gen Agent Trust Hub on Jul 13, 2026

Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses sensitive information in the ~/.cursor/projects directory, which contains project transcripts and interaction history.
  • [COMMAND_EXECUTION]: The instructions direct the agent to execute shell commands using pnpm (e.g., pnpm harvest:all, pnpm seed-manifest) within the local repository.
  • [PROMPT_INJECTION]: The skill processes untrusted local data, introducing a risk of indirect prompt injection. Ingestion points: Local .jsonl transcript files in ~/.cursor/projects. Boundary markers: No delimiters are specified to isolate processed content. Capability inventory: File system access and shell command execution via local scripts. Sanitization: No content validation or sanitization is mentioned.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 13, 2026, 08:47 AM
Security Audit — agent-trust-hub — cursor-transcript-harvest