read-claude-session

Pass

Audited by Gen Agent Trust Hub on Oct 25, 2026

Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill is designed to read and process local Claude Code session transcripts, which may contain sensitive information.\n
  • Evidence: The script scripts/session_export.py accesses the ~/.claude/projects/ directory to locate and read .jsonl transcript files.\n
  • Context: This behavior is essential for the skill's primary function. The documentation includes explicit warnings to treat transcripts as sensitive data and to avoid reproducing credentials or personal content.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external session transcripts which could contain malicious instructions designed to influence the agent's output.\n
  • Ingestion points: The scripts/session_export.py script reads .jsonl files from the local filesystem.\n
  • Boundary markers: The exporter uses Markdown headers (## User, ## Assistant) and horizontal rules (---) to separate different parts of the conversation.\n
  • Capability inventory: The script is limited to file reading and writing. The skill instructions suggest using standard text-processing utilities like wc and rg on the output.\n
  • Sanitization: The script employs regular expressions to strip system-specific tags such as <system-reminder> and <channel> from the processed text.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 25, 2026, 07:45 AM