gsd-path-inspect
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill implements a 'Project Runtime' mechanism (documented in 'references/runtime-selection.md'). When a '.gsd-path/runtime.json' file is present in the repository being analyzed, the agent is instructed to execute a 'launcher' script ('.gsd-path/status_runtime.py') from that repository and use its output to resolve other helper tools. This allows the target codebase to execute arbitrary Python code within the agent's environment.
- [COMMAND_EXECUTION]: The 'scripts/loop_run.py' utility is designed to execute arbitrary shell commands that are parsed from a 'LOOP.md' specification file in the project directory. This capability is used to automate fix-and-verify loops but provides a path for command execution driven by project-level configuration.
- [INDIRECT_PROMPT_INJECTION]: The workflow relies heavily on the contents of Markdown files found in the target repository, such as 'INTENT.md', 'PLAN.md', and 'ROADMAP.md'. These files define the agent's tasks, success criteria, and routing logic (Category 8). While the skill includes strict schema validation and boundary markers (e.g., in 'scripts/check_handoffs.py'), a malicious repository could craft these files to steer the agent's planning phase toward unauthorized actions.
- [COMMAND_EXECUTION]: Multiple scripts, including 'scripts/integration.py', use the GitHub CLI ('gh') and Git to perform network operations, including pushing branches and managing pull requests on 'github.com'. These actions are consistent with the skill's primary purpose but represent high-capability tool usage.
Audit Metadata