The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests Structured Threat Information Expression (STIX) objects from external TAXII servers, representing a surface for instructions embedded in external data.
Ingestion points: The TAXIIConsumer.poll_collection method in SKILL.md and the scripts/agent.py audit tool ingest objects from potentially untrusted third-party TAXII server endpoints.
Boundary markers: No delimiters or instructions to ignore embedded commands are present when processing the threat intelligence bundles.
Capability inventory: The skill includes logic to push ingested data to external platforms like Splunk and Elasticsearch via network requests and writes metadata to local JSON files (taxii_data.json).
Sanitization: Data fields such as object descriptions and names are extracted and forwarded to SIEM platforms without sanitization or validation of the content.
[EXTERNAL_DOWNLOADS]: The skill configuration and documentation instruct the installation of various Python packages from public registries.
Evidence: The docker-compose.yml file in SKILL.md executes pip install medallion at container startup, and references/api-reference.md lists commands to install opentaxii, taxii2-client, stix2, and requests.
Sources: The skill targets standard, well-known libraries from the cybersecurity domain (e.g., OASIS Open CTI implementations).
[COMMAND_EXECUTION]: Deployment instructions rely on shell command execution to manage dependencies and launch server instances.
Evidence: The command field in the Docker service definition in SKILL.md uses bash -c to chain package installation and service execution.

implementing-taxii-server-with-opentaxii