The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/check_walkthrough.py executes git commands using subprocess.run to verify source citations. It includes several security best practices to mitigate command injection:
It passes arguments as a list rather than a single string.
It uses the --end-of-options flag in git rev-parse to prevent argument injection.
Git revisions are strictly validated using a regex that only allows the literal 'WORKTREE' or hex strings (git SHAs), preventing arbitrary command flags from being passed.
File paths are checked to ensure they are repository-relative and do not contain path traversal characters (.. or ).
[INDIRECT_PROMPT_INJECTION]: The skill operates by reading and analyzing user-provided source code and task context, which serves as an attack surface for indirect prompt injection.
Ingestion points: The agent reads project instructions, task context, and source files from the target repository (SKILL.md, references/pipeline.md).
Boundary markers: The SKILL.md instructions explicitly state that retrieved text must be treated as evidence and cannot change scope, permissions, or approvals.
Capability inventory: The skill has the ability to execute local Git commands and write Markdown walkthrough files to specific directories.
Sanitization: The associated checker script validates that excerpts and citations match the source repository using strict path and revision validation, reducing the risk of a malicious file misleading the verification process.

code-walkthrough