The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted creator materials and metadata files (JSON, YAML, text) to generate rights reports. Maliciously crafted metadata could attempt to influence the agent's summary or instructions during the packaging workflow.
Ingestion points: In scripts/create_transfer_package.py, the skill reads local files and metadata files (e.g., metadata.json, metadata.yaml) from user-provided source directories.
Boundary markers: The skill instructions emphasize labeling uncertain facts as "unknown" or "needs confirmation" and explicitly flag generated reports as "Private draft" to signal they require human review.
Capability inventory: The skill is limited to local file system operations, including listing files, reading contents, calculating SHA-256 hashes, and writing Markdown/JSON reports. It contains no network access or subprocess execution capabilities.
Sanitization: The create_transfer_package.py script implements a comprehensive denylist of sensitive directory names (.ssh, .aws, .git, .env, etc.) and filenames (credentials, id_rsa, .env.local) to prevent accidental data exposure. It also uses yaml.safe_load for secure parsing of YAML metadata.

suede-rights-passport