The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with a local utility named low-data to manage Git and GitHub network policies. It includes explicit instructions to resolve the command path using command -v and verify its local source before execution. Crucially, it forbids the agent from downloading, replacing, or installing the command if it is missing, mitigating potential remote code execution risks.
[INDIRECT_PROMPT_INJECTION]: The agent is instructed to ingest and interpret the output of low-data status and low-data active. While this represents an ingestion point for external data, the risk is minimized by the requirement that the command be a trusted local prerequisite. The skill also specifies strict exit code checks (exit 0 for active, exit 1 for inactive) rather than relying solely on natural language output parsing.
Ingestion points: SKILL.md
Reading output from low-data status and low-data active commands.
Boundary markers: Absent for text output, but the workflow utilizes explicit exit code verification logic to determine state.
Capability inventory: Shell command execution via env and low-data for status checks and mode switching.
Sanitization: The skill relies on exit code semantics rather than sanitizing the string output of the tool.

github-airplane-mode