Skip to main content

Security

Path Security​

All skill names and file paths are validated before writing to disk.

Name Sanitization​

The sanitizeName() function enforces the agentskills.io spec:

  • Unicode NFKD normalization (e.g., cafe-resume from café-résumé)
  • Lowercased, kebab-case only
  • Max 64 characters
  • Matches: ^[a-z0-9]([a-z0-9-]*[a-z0-9])?$
  • Null bytes (\x00) stripped
  • Path traversal sequences (../) blocked

Path Validation​

isPathSafe() resolves symlinks and verifies the target is within the allowed root directory. safeWrite() combines validation and atomic writing.

isSafeSymlink() verifies symlink targets don't escape the allowed directory boundary.

Lock File Integrity​

Skills are tracked in ~/.claude/auto-skill/skills.lock.json with SHA-256 content hashes. Verify integrity:

auto-skill lock verify

The lock file uses atomic writes (temp file + fs.renameSync()) to prevent corruption.

Spec Compliance​

The validateSkillMd() function checks generated skills against the agentskills.io specification:

  • Name format and length
  • Description under 1024 characters
  • allowed-tools as YAML list (not comma-separated string)
  • version field present
  • Valid YAML frontmatter