review-and-refactor
Fail
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: CRITICALINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon instructions provided in external repository files.
- Ingestion points: The agent is instructed to review all coding guidelines in
.github/instructions/*.mdand.github/copilot-instructions.mdbefore performing tasks. - Boundary markers: There are no boundary markers or instructions to treat the content of these files as data rather than trusted instructions, which could allow those files to override agent behavior.
- Capability inventory: The skill allows for code refactoring (file system write) and task completion verification (likely involving shell execution for tests).
- Sanitization: No sanitization or validation steps are defined for the content read from these external guideline files.
- Ingestion points: The agent is instructed to review all coding guidelines in
- [COMMAND_EXECUTION]: The task requirement to "ensure they [tests] are still passing" requires the agent to execute shell commands. When combined with the ingestion of untrusted instructions from the repository, this capability could be abused to execute malicious commands if the instruction files are compromised.
Recommendations
- CRITICAL: 1 file(s) identified as malware by FileRep - DO NOT USE
Audit Metadata