paseo-rescue

Pass

Audited by Gen Agent Trust Hub on Sep 11, 2026

Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill instructs the agent to access sensitive application data within the user's home directory, specifically targeting ~/.paseo/agents/ and ~/.claude/projects/. It explicitly retrieves persistence.nativeHandle, which acts as a session identifier, and reads raw .jsonl transcripts. While this is the primary function for "rescuing" a session, it involves handling potentially sensitive project histories and authentication artifacts.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external, historical agent sessions which may contain untrusted content.
  • Ingestion points: Content is retrieved from the get_agent_activity tool or directly from local transcript files in the file system.
  • Boundary markers: The instructions require the agent to generate a "Takeover Summary" and wait for user confirmation before taking any further action, which acts as a manual checkpoint.
  • Capability inventory: The skill utilizes agent listing, activity retrieval, and file system read capabilities.
  • Sanitization: The agent is instructed to "clean" the timeline by removing noise, truncated JSON inputs, and language artifacts before presenting the summary to the user.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 11, 2026, 07:20 AM
Security Audit — agent-trust-hub — paseo-rescue