The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements dynamic regular expression compilation for flag targeting rules.
Evidence: new RegExp(v).test(attributeValue) in FeatureFlagService.evaluateTargeting (SKILL.md).
Risk: If flag configuration values are sourced from untrusted inputs, it creates a vulnerability to ReDoS attacks.
[DATA_EXFILTRATION]: Implementation patterns perform network requests to computed endpoints retrieved from configuration.
Evidence: fetch(variant.healthEndpoint) in InfrastructureFlagService and fetch(modelVariant.endpoint) in generateResponse (SKILL.md).
Risk: Potential SSRF surface if the backend configuration store (e.g., Redis) is compromised or accepts untrusted data.
[PROMPT_INJECTION]: The skill processes untrusted user attributes that can influence agent behavior through conditional feature gates.
Ingestion points: EvaluationContext in SKILL.md.
Boundary markers: Absent; user attributes are directly converted to strings for rule evaluation.
Capability inventory: fetch() for network requests to external model and health endpoints; redis.set() for managing state.
Sanitization: Absent; input attributes are processed without escaping or strict validation against a schema.

feature-flags