The Agent Skills Directory

[SAFE]: The skill incorporates extensive security features designed to prevent unauthorized system access or data corruption:
Transaction Safety: Uses a robust file transaction model that stages changes in temporary files and performs an atomic rename into place, with automated rollback and cleanup on failure.
Integrity Enforcement: Maintains a hash-based state registry to track managed files; it detects manual modifications and blocks updates to prevent overwriting user-owned data.
Path Hardening: Strictly validates all filesystem operations to ensure they stay within the target repository, refuses to follow symbolic links, and checks for symlink ancestors to prevent directory traversal.
Approval Control: Every mutating operation requires the user to review a JSON plan and approve it by providing a single-use SHA-256 hash, ensuring no background writes occur without explicit consent.
[COMMAND_EXECUTION]: The skill orchestrates various internal tools and scripts. It uses spawnSync with shell: false for all process invocations, preventing command injection vulnerabilities. The execution environment for sub-processes is limited to a minimal whitelist of necessary system environment variables.
[INDIRECT_PROMPT_INJECTION]: The generator processes skill names and descriptions from directories within the repository to produce marketplace manifests. It employs strict regex validation for identifiers and applies comprehensive Markdown escaping to metadata before interpolation to prevent content from influencing agent behavior during installation or catalog browsing.

create-skills-repo