show-me-your-work

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill includes and utilizes a local bash script scripts/log.sh to append decision entries to a TSV file. The script performs directory creation via mkdir -p and file appending using printf and the >> operator.
  • [INDIRECT_PROMPT_INJECTION]: The skill instructions define an audit and cross-model review process where agents and subagents ingest session transcripts and logs to verify work.
  • Ingestion points: The audit phase involves reading the host's conversation transcript and the decisions.tsv file, both of which contain strings generated during the run or potentially derived from user input.
  • Boundary markers: The log uses TSV delimiters. While instructions tell the agent to "check the log told the truth," there are no explicit boundary markers or instructions for the subagent to disregard potential instructions embedded within the transcript text being reviewed.
  • Capability inventory: The skill has the ability to write to the file system (scripts/log.sh), retrieve session history (/recall), and spawn external subagents for review (/swarm).
  • Sanitization: The scripts/log.sh script implements defensive sanitization for spreadsheet applications by stripping newlines/tabs and prefixing potential formula characters (=, +, -, @) with a single quote to prevent formula injection attacks when the log is opened in tools like Excel or Google Sheets.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 09:16 AM