The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, including various file formats (PDF, PPT, Excel, Word) and URLs.
Ingestion points: Files and URLs are passed via the target parameter to the security-scan stage.
Boundary markers: The instructions do not explicitly define strict delimiters or sanitization protocols for the content being scanned, which could allow malicious instructions embedded in documents to influence downstream agent actions.
Capability inventory: The skill has the capability to write scan reports to the local file system (examples/snse_survey/skill_outputs/security_scan.json) and influence the execution flow through its verdict (pass/reject).
Sanitization: No explicit sanitization or filtering of external content is visible in the provided orchestration script.
[COMMAND_EXECUTION]: The execution script scripts/run_security_scan.py triggers a runtime stage security-scan via the AgentSession.run_stage method. This invokes internal logic to perform the scanning tasks.
[EXTERNAL_DOWNLOADS]: The skill documentation lists python-magic as a dependency. This is a standard, well-known library for identifying file types and does not represent an unverified or high-risk dependency.

antinet-security-scan