The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external specification data to drive agent actions, creating a surface where malicious instructions could be embedded in user-provided documents.- Ingestion points: Specification documents (templates/SPEC.md) and task plans (templates/TASK-PLAN.md) are the primary data inputs for the agent during the factory pipeline.- Boundary markers: The methodology requires four distinct review gates (Gate 1-4) using templates/REVIEW.md, and specific quality checks outlined in references/spec-quality-gates.md to identify and block ambiguous or non-testable instructions.- Capability inventory: The associated scripts (scripts/spec-quality-check.sh, scripts/spec-to-tasks.sh) are limited to local text processing using grep and do not provide the agent with network or elevated system capabilities.- Sanitization: The methodology encourages the use of structured machine-readable formats like Gherkin and OpenAPI (references/format-translation.md), which limit the success of unstructured instructional overrides compared to natural language prompts.

spec-driven-development