gm-finish
Pass
Audited by Gen Agent Trust Hub on Oct 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill performs automated file system maintenance by pruning old JSON response files from the
.gm/exec-spool/out/directory. - [COMMAND_EXECUTION]: The skill executes network operations via git commands, specifically committing and pushing logs and archives to a remote repository.
- [INDIRECT_PROMPT_INJECTION]: The skill processes 'PRD rows' as external input, creating a potential surface for indirect prompt injection.
- Ingestion points: The skill reads data from open PRD rows, likely stored in
.gm/prd-log.jsonlor similar project files. - Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between row data and execution instructions.
- Capability inventory: The skill possesses the capability to delete files, perform git push operations, and load/execute other subagent skills (
gm-orient). - Sanitization: No validation or sanitization of the project row content is described before the data is processed or used to load subagents.
Audit Metadata