The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted email content from external .eml files, creating a potential attack surface for indirect prompt injection.
Ingestion points: scripts/agent.py and scripts/process.py read content from files provided via the --eml-file argument.
Boundary markers: The scripts do not use explicit delimiters or "ignore instructions" warnings when processing the untrusted email body or headers.
Capability inventory: The scripts/agent.py script includes capabilities for outbound network requests (requests.get, requests.post) and local file system writes.
Sanitization: The implementation relies on regular expressions for indicator extraction but lacks specific sanitization to prevent malicious content within the email from influencing the logic of an agent interpreting the output.
[DATA_EXFILTRATION]: The scripts/agent.py script performs network operations to external endpoints.
The script makes requests to VirusTotal for URL reputation checks, which is a well-known security service.
The script includes functionality to POST incident data to a user-supplied ticketing_url. While this is core functionality for the workflow, the ability to transmit data to an arbitrary external URL using a bearer token constitutes a low-risk exfiltration surface.

building-phishing-reporting-button-workflow