The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.check_call to execute a shell command for package management. Specifically, it calls pip install PyGithub if the library is missing from the environment. The command is constructed as a static list of arguments, which prevents command injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The script scripts/github_tool.py dynamically downloads and installs the PyGithub package from the official Python Package Index (PyPI) at runtime if it is not already installed. This is a common pattern for managing dependencies in standalone scripts but involves fetching external code.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the processing of untrusted data from GitHub.
Ingestion points: The list_issues and get_issue actions in scripts/github_tool.py fetch titles, bodies, and comments from external GitHub repositories.
Boundary markers: The skill does not implement boundary markers or instructions to the agent to ignore embedded commands within the retrieved issue content.
Capability inventory: The skill has write capabilities, including create_issue and comment_issue, which could be abused if an agent follows instructions found within a retrieved issue.
Sanitization: The content retrieved from GitHub is returned as raw strings within JSON without filtering or sanitization.

github-manager