om-sync-merged-pr-issues
Pass
Audited by Gen Agent Trust Hub on Jul 14, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill includes an explicit 'Untrusted content boundary' section designed to prevent instructions embedded in repository data (such as PR descriptions or issue comments) from being followed. The agent is instructed to treat all repository content as data and to report any detected directives like 'ignore previous instructions' as suspected prompt injections. The deterministic flag for this pattern is a false positive triggered by these safety instructions. Furthermore, the skill processes external data through defined ingestion points with explicit boundary markers to mitigate indirect prompt injection risks.
- [COMMAND_EXECUTION]: The skill uses shell commands for system tasks like timestamp generation. It implements defense-in-depth for command safety by requiring strict regex validation (e.g., ensuring issue IDs are numeric) and double-quoting all variables before they are interpolated into shell strings, effectively preventing command injection from malformed or malicious repository metadata.
Audit Metadata