show-me-your-work
Pass
Audited by Gen Agent Trust Hub on Aug 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes and logs potentially untrusted data (like PR numbers, commit messages, or filenames) into a decision trail. While the shell script
scripts/log.shproperly sanitizes these for spreadsheet formula injection (CSV injection), the agent reading these logs later is still subject to indirect prompt injection if the logs contain adversarial instructions. - [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read chat transcripts located at
~/.claude/projects/to verify log accuracy. These transcripts contain external input (human messages or tool outputs) that could contain malicious instructions designed to influence the 'self-audit' or 'cross-model review' phases. - [COMMAND_EXECUTION]: The skill uses a bash script
scripts/log.shto append data to local files. While the script is locally provided and uses standard utilities (date,printf,tr), it is invoked with user-supplied and agent-generated arguments. The sanitization in the script focuses on spreadsheet safety rather than shell safety, thoughprintfusage here is generally safe against typical command injection when handled by the agent environment.
Audit Metadata