The Agent Skills Directory

[SAFE]: The skill implements a restricted file creation workflow that limits operations to specific sanctioned directories (.agents/skills/ and .agents/skill-drafts/) within the current workspace.
[SAFE]: Robust validation is performed on user-supplied identifiers (skill-id), explicitly blocking path traversal sequences (.., /), absolute paths, and Windows reserved filenames (e.g., con, prn, nul).
[SAFE]: The execution of shell commands is limited to a local validation tool (muse skills validate) using the strictly sanitized directory path, effectively mitigating command injection risks.
[SAFE]: The skill follows the principle of least privilege by refusing to write to foreign harness roots or managed configuration directories directly, instead providing the user with a specific command to perform managed installations.
[SAFE]: No evidence of network exfiltration, obfuscation, or unauthorized credential access was found in the instructions or defined behavior.

create-skill