The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs a shell command using user-provided input for the issue title. While the issue body is protected by a quoted heredoc, the title placeholder (<title>) is interpolated within double quotes. This presents a surface for indirect prompt injection where a maliciously crafted title could lead to unintended command execution.\n
Ingestion points: User-provided refactoring titles and descriptions within the process defined in SKILL.md.\n
Boundary markers: Quoted heredoc markers (<<'EOF') are used for the body, providing protection against injection in that section, but markers are absent for the title placeholder.\n
Capability inventory: Executes gh issue create and gh auth status commands.\n
Sanitization: No explicit sanitization or escaping mechanisms are specified for the user-provided strings before they are interpolated into the shell command.\n- [DATA_EXFILTRATION]: The skill transmits data to GitHub, which is a well-known and trusted service. This behavior is consistent with the skill's stated purpose and does not involve accessing sensitive local data like credentials or SSH keys, making it a standard functional component.

refactor-plan