The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/card.py utility executes the cosign command-line tool to perform Sigstore-based signing and verification of artifacts. These operations are performed using subprocess.run with list-based arguments, a secure implementation that avoids shell interpolation and prevents command injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill scans directories and extracts metadata from Markdown files, creating an attack surface for indirect prompt injection. Ingestion points: Metadata is read from SKILL.md and other files within a bundle directory by scripts/card.py. Boundary markers: Extracted metadata is placed into the CARD.md frontmatter; while specific delimiters are absent, the script uses json.dumps to sanitize values against YAML structure breakage. Capability inventory: The skill performs filesystem writes and executes external signing tools. Sanitization: The script's reliance on list-based subprocess calls prevents malicious metadata from triggering arbitrary command execution.
[SAFE]: The skill demonstrates strong security awareness by implementing restrictive filesystem permissions (0600) for generated private keys and providing a graded trust model that distinguishes between cryptographic integrity and runtime behavior. It also includes a capability manifest (permissions.yaml) that explicitly declares its required permissions.

trust-card