show-me-your-work
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill directs the agent to read sensitive platform-specific session logs from the filesystem to verify its audit trail. Evidence includes instructions in SKILL.md targeting paths such as ~/.codex/sessions/* and ~/.claude/projects/*. This access is purposeful for auditing but involves sensitive user data.
- [COMMAND_EXECUTION]: The skill executes a bundled shell script (scripts/log.sh) to maintain the decision log. The script performs directory creation and file appending. Notably, it includes a sanitization step to prevent spreadsheet formula injection by prefixing characters like '=', '+', '-', and '@' with a single quote.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data (session transcripts) that may contain untrusted content. Ingestion occurs through platform log files and CLI outputs. While the skill includes input sanitization for its log output, the reading of transcripts presents a surface for potential instruction injection, though restricted by the skill's specific auditing instructions.
Audit Metadata