The Agent Skills Directory

[SAFE]: The skill implements a comprehensive safety model. It restricts its own execution to tasks that have already been human-reviewed through a 'ready' handoff process.
[SAFE]: Detailed 'Hard Stops' are defined in both SKILL.md and references/approval-policy.md, specifically forbidding the agent from handling credentials, secrets, or security-sensitive decisions automatically.
[SAFE]: The skill uses localized instructions (Korean) for legitimate tooling guidance (e.g., seeding skeletons with wt scaffold), which is appropriate for the author's context and does not involve obfuscation or malicious homoglyphs.
[SAFE]: No suspicious network operations, remote code execution, or data exfiltration patterns were detected. All described file operations pertain to the project's own execution state (e.g., .wt/ directory).

autopilot