The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/exception_creep.py utilizes subprocess.run to invoke the git command for diff analysis. The implementation correctly passes arguments as a list and avoids the use of a shell, which follows security best practices for executing system commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of code diffs and source files to generate audit findings. While this creates a theoretical surface for indirect prompt injection via maliciously crafted code comments, the risk is minimal as the skill lacks dangerous capabilities like network access or persistent file modifications.
Ingestion points: scripts/exception_creep.py processes diff data from git or patch files; scripts/gate_selftest.py reads local Python source code.
Boundary markers: Findings are presented as structured text or JSON but do not explicitly wrap content in protective delimiters.
Capability inventory: The skill is restricted to local file reading and executing local git commands; it lacks network or file-write capabilities.
Sanitization: Content from analyzed files is included in findings with basic truncation and whitespace stripping.

drift-to-gate