The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs file read and write operations on specific markdown files (.copilot/goals.md and ~/.copilot/goals.md). These operations are restricted to the intended purpose of the skill and do not involve arbitrary command execution or access to sensitive system files.
[PROMPT_INJECTION]: The skill processes human-editable files, which theoretically introduces a surface for indirect instructions. However, the instructions explicitly require the agent to surface active goals to the user and wait for direction before acting, effectively using a human-in-the-loop validation to prevent automatic execution of untrusted content.
[CREDENTIALS_UNSAFE]: The skill documentation includes a specific anti-pattern warning against storing secrets, credentials, or personally identifiable information (PII) in the goals files, promoting secure usage of the tool.

long-running-goals