The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The bootstrap.py script downloads a ZIP archive from GitHub. The download process includes strict URL validation and SHA-256 checksum verification before extraction or execution.
[COMMAND_EXECUTION]: The skill frequently uses the subprocess module to execute a native CLI tool (printcraft-cli) and an OCR engine (tesseract). These executions are restricted to specific subcommands and paths validated by the skill's scripts.
[DYNAMIC_EXECUTION]: Scripts such as cli.py, commands.py, and command_gateway.py utilize importlib.util to dynamically load internal Python modules from the skill's own directory. This is used for modularizing the skill's logic.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from PDF files and tool catalogs.
Ingestion points: PDF content is read via the text command in verification.py, and JSON catalogs are ingested in command_gateway.py.
Boundary markers: The skill uses structured JSON schemas to validate tool inputs and outputs, though it does not use explicit natural language boundary markers for prompt injection protection.
Capability inventory: Uses subprocess.run and subprocess.Popen to interact with the native binary and the shell.
Sanitization: execution.py implements a redaction mechanism that scans outputs for common sensitive patterns (e.g., password, api_key, token) and replaces them with [REDACTED] before returning data to the agent.

pdfcraft-cli