The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an ingestion pipeline for external threat intelligence feeds (OSINT) from sources like URLhaus and Feodo Tracker. This creates an attack surface where malicious actors could embed prompt injection instructions within threat data to influence the agent's behavior during analysis.
Ingestion points: SKILL.md (Workflow Step 2), scripts/process.py (TIPManager.configure_feeds).
Boundary markers: No specific delimiters or "ignore instructions" warnings are implemented around the ingested feed data.
Capability inventory: The skill possesses extensive API write capabilities to MISP, OpenCTI, TheHive, and Cortex (scripts/process.py), and generates STIX indicators (scripts/agent.py).
Sanitization: Data is processed using standard library parsers (pymisp, pycti) without additional sanitization for natural language instructions.
[CREDENTIALS_UNSAFE]: The Docker Compose configuration in SKILL.md includes hardcoded default credentials for the OpenCTI platform, such as APP__ADMIN__PASSWORD=TIPAdminPassword and APP__ADMIN__TOKEN=tip-opencti-token-uuid. While presented as examples for a development environment, these should be managed via secure secrets in production.
[EXTERNAL_DOWNLOADS]: The platform deployment process fetches container images from well-known and official repositories, including the GitHub Container Registry (ghcr.io/misp), Elastic's registry, and Docker Hub (Redis, RabbitMQ, Minio). These downloads are associated with established security software and infrastructure providers.
[DATA_EXFILTRATION]: The management scripts (scripts/process.py and SKILL.md code snippets) configure connections to MISP using ssl=False. This disables certificate verification, which exposes potentially sensitive threat intelligence data to man-in-the-middle attacks if used outside of a strictly isolated environment.

building-threat-intelligence-platform