transcript-analysis
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides structured methods for auditing and analyzing session transcripts stored in ~/.claude/projects/. This access is appropriate for a forensics and performance analysis tool. The use of MotherDuck and DuckDB for querying local logs is a standard practice and no malicious code or exfiltration attempts were found.
- [INDIRECT_PROMPT_INJECTION]: The skill processes session transcripts that contain untrusted data from previous human prompts and tool outputs. Although this presents a theoretical attack surface where an agent could be influenced by instructions found within the logs, it is an inherent risk of log analysis and is categorized as low risk. 1. Ingestion points: JSONL files in ~/.claude/projects/. 2. Boundary markers: Data is structured as JSON, though no explicit 'ignore instructions' markers are used during analysis. 3. Capability inventory: Execution of SQL queries via kaizen-duckdb and file-write access to .planning/kaizen/. 4. Sanitization: Content is parsed as JSON; no additional sanitization is specified for the text fields before they are presented to the agent for analysis.
Audit Metadata