show-me-your-work
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes interaction history from the
agent-transcripts/directory to verify the accuracy of the decision log. - Ingestion points: The agent reads local transcript files (e.g., in
agent-transcripts/) which capture all previous turns and external tool outputs. - Boundary markers: Instructions explicitly limit the scope to the current run's transcript and warn the agent not to access unrelated project chats in
~/.cursor/projects/*/. - Capability inventory: The skill uses
scripts/log.shto write and append data to local TSV files. - Sanitization: The logging script (
scripts/log.sh) implements aclean()function that strips tabs and newlines and prefixes formula-triggering characters (=,+,-,@) with a quote to prevent CSV/Formula injection when the log is viewed in spreadsheet software. - [COMMAND_EXECUTION]: The skill relies on a local bash script (
scripts/log.sh) to manage the decision log. The script performs standard filesystem operations such as directory creation and timestamped file appending using securely quoted variables.
Audit Metadata