The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/package_sync.py uses subprocess.run to execute git status commands. This allows the skill to identify which skill directories have changed and need to be repackaged. The implementation follows security best practices by passing arguments as a list rather than a shell string, effectively preventing command injection.
[DATA_EXFILTRATION]: The skill's packaging logic in scripts/package_sync.py creates ZIP archives of entire skill directories. As specified in SKILL.md, this process includes all files in the directory, such as auth.json, if they are present. While this is consistent with the skill's intended purpose of preparing packages for upload, it creates a potential data exposure surface if users include sensitive credentials in their skill folders.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it processes output from external system tools. 1. Ingestion points: scripts/package_sync.py ingests file names and change summaries from the output of git status. 2. Boundary markers: The script does not utilize specific delimiters to separate these filenames from the agent's internal instructions. 3. Capability inventory: The skill possesses capabilities for file modification, directory packaging, and system command execution. 4. Sanitization: There is no explicit validation or filtering performed on the filenames retrieved from the system. However, the risk is limited as the data source is the local Git environment.

agent-package-sync