The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to process external threat intelligence from arbitrary URLs or text inputs, creating a surface where instructions embedded in that data could attempt to influence the agent's behavior. This risk is addressed through a structured security gate and a defensive directive.
Ingestion points: External URLs for threat research blogs and raw intelligence text (Workflow 2, Step 1).
Boundary markers: The skill includes a 'Prompt Injection Defense Directive' and explicit instructions to scan for adversarial patterns such as 'ignore instructions'.
Capability inventory: The skill utilizes tools to create, validate, and test YARA-L detection rules, which could potentially be influenced by untrusted data if not for the built-in safeguards.
Sanitization: Workflow 2, Step 1 details a multi-step process for stripping HTML elements and scanning for adversarial strings before data is processed.
External Data Retrieval: As part of its core functionality, the skill retrieves content from external sources such as threat blogs and CVE advisories. This behavior is necessary for threat detection evaluation, and the skill implements procedures to clean and normalize this data before it is utilized.
Controlled Environment Modification: The skill possesses the capability to deploy detection rules into the production environment. To ensure integrity and oversight, the instructions require explicit user authorization before any permanent changes are committed via the create_rule tool.

secops-detection-engineering