The Agent Skills Directory

[COMMAND_EXECUTION]: The provided helper script scripts/docmaint utilizes subprocess.run to execute git commands. These calls are used for repository introspection, including listing files (ls-files), checking commit history (rev-parse), and identifying changed files (diff). The commands are invoked using argument lists, which prevents shell injection vulnerabilities.
[PROMPT_INJECTION]: The skill is designed to process external documentation files, which constitutes a potential surface for indirect prompt injection (Category 8).
Ingestion points: The agent is instructed to read markdown files from the target repository, and the scripts/docmaint script reads these files to scan for terminology violations.
Boundary markers: There are no explicit XML-style or unique string delimiters used when the agent reads the documentation content.
Capability inventory: The skill has the ability to write to files and execute git commands through the docmaint utility.
Sanitization: The skill does not implement automated sanitization of the markdown content.
Mitigation: The skill includes robust procedural guardrails, such as mandatory classification gates (evergreen vs. point-in-time), an explicit rubric that limits auto-fixes to mechanical changes, and requirements for independent verification and human confirmation, which effectively mitigate the risk of following instructions embedded within the data.
[SAFE]: No evidence of data exfiltration, hardcoded credentials, malicious persistence, or obfuscation was found. The Python scripts rely exclusively on the standard library and perform operations consistent with the skill's stated purpose of documentation maintenance.

maintaining-documentation