The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the printcraft-cli binary from GitHub (via the storytold/pdfcraft or full-aigc-skills/pdfcraft-skills repositories). This process is secured by hardcoded SHA-256 checksums in runtime.lock.json, which are verified before any files are extracted or executed. The scripts/bootstrap.py script further enforces HTTPS and sanitizes the archive content to prevent directory traversal attacks.
[COMMAND_EXECUTION]: The skill utilizes subprocess.run and subprocess.Popen to call its verified CLI tool and local Python scripts. It avoids shell injection by passing arguments as lists and validates task configurations against a strict JSON schema in scripts/command_gateway.py.
[CREDENTIALS_SAFE]: A specialized redaction mechanism in scripts/execution.py scans for sensitive keywords such as api_key, token, and password in task parameters and command outputs, replacing them with [REDACTED] in publicly visible task receipts and logs.
[INDIRECT_PROMPT_INJECTION]: While the skill processes external PDF content, it mitigates potential injection risks by using a specialized binary tool for data extraction, performing deterministic integrity checks on outputs, and requiring explicit visual review steps for finalized artifacts.

pdfcraft-use