The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/inspect_pdf.py uses subprocess.run to call the pdfinfo utility. The execution is handled securely by passing arguments as a list and using shutil.which to resolve the executable path, preventing shell injection vulnerabilities.
[PROMPT_INJECTION]: The skill contains comprehensive instructions to identify and mitigate indirect prompt injection risks. It specifically directs the agent to scan for and treat active PDF content—such as /JavaScript, /OpenAction, and /Launch markers—as inert hazards, preventing them from being executed or influencing the agent's behavior.
[DATA_EXFILTRATION]: The skill's instructions strictly enforce local processing of confidential documents and forbid the transmission of data to external services without explicit user consent. The provided source code contains no network-reaching functionality or evidence of data exfiltration.
[CREDENTIALS_UNSAFE]: The workflow includes explicit prohibitions against requesting sensitive credentials like private keys in the chat and mandates the use of secure channels for any necessary passwords, adhering to industry best practices for secret management.
[COMMAND_EXECUTION]: The helper script scripts/inspect_pdf.py implements a secure file-writing mechanism (atomic_write_text) that uses temporary files and rigorous path validation to ensure that output files do not unintentionally overwrite input documents or sensitive system files.

pdf-processing