setup

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill transmits session telemetry metadata (such as token counts, build times, and tool usage) to the vendor's infrastructure at promptbook.gg.
  • Evidence: Multiple steps (1, 3, 4, 6) use curl to send data to https://promptbook.gg.
  • Context: This is the primary function of the skill, and the data scope is explicitly documented in the privacy note to exclude source code or prompt contents.
  • [INDIRECT_PROMPT_INJECTION]: The skill scans local agent session history files to backfill build data, which involves processing previously recorded interaction logs.
  • Ingestion points: ~/.claude/projects/*.jsonl files (Step 8).
  • Boundary markers: None specified in the instructions.
  • Capability inventory: find, node execution, and curl network operations.
  • Sanitization: None specified in the shell-level interaction.
  • [COMMAND_EXECUTION]: The skill executes a bundled JavaScript file (backfill-history.js) using Node.js to process history files.
  • Evidence: Step 8 invokes node on a file path discovered locally via the find command.
  • Context: The script is part of the local plugin installation and is not downloaded from an external source at runtime.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 03:47 AM
Security Audit — agent-trust-hub — setup