The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes a maintenance script (scripts/update_secret_rules.py) that fetches updated regex patterns and license information from the official Gitleaks GitHub repository. This facilitates keeping the redaction rules current by targeting a well-known and trusted security project.
[COMMAND_EXECUTION]: The redaction logic invokes local system binaries, specifically gitleaks for secret scanning and tesseract for OCR. The implementation utilizes subprocess.run safely by passing arguments as lists and resolving executable paths via shutil.which, preventing common shell injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: Since the skill processes untrusted documents and images, it is exposed to potential indirect prompt injection. The skill implements robust defenses by providing the agent with explicit instructions to treat document content as untrusted and by performing irreversible redaction on the data before it is presented in higher-level contexts.
[DATA_EXFILTRATION]: Communication with the Replicate API is included for optional cosmetic image editing. This network activity is purpose-specific, restricted to artifacts that have already undergone a redaction pass, and is protected by mandatory user consent flags to prevent accidental data transmission.

anonymizer