show-me-your-work

Pass

Audited by Gen Agent Trust Hub on Nov 18, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill requires the agent to audit project transcripts and decision logs, which may contain untrusted data from external sources like pull request descriptions, commit messages, or tool outputs.
  • Ingestion points: The skill reads files from the agent-transcripts/ directory to verify the accuracy of the decision log.
  • Boundary markers: While the log uses TSV formatting, there are no specific prompt delimiters defined for the cross-model subagent review process.
  • Capability inventory: The skill allows writing to the filesystem, reading session transcripts, and spawning subagents for review.
  • Sanitization: The skill includes a bash script (scripts/log.sh) that sanitizes input by stripping control characters (tabs, newlines) and prefixing formula-triggering characters (=, +, -, @) with a single quote to prevent injection when the log is opened in spreadsheet software.
  • [COMMAND_EXECUTION]: The skill utilizes a local bash script (scripts/log.sh) for logging. The script is designed with security best practices, including proper variable quoting and input sanitization, to safely handle data and manage directories via mkdir -p.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The instructions explicitly warn the agent against using broad glob patterns (e.g., ~/.cursor/projects/*/) to avoid accidental exposure of private chat history from unrelated projects, demonstrating a privacy-conscious design.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 18, 2026, 01:48 PM