The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided evidence and external research data, which constitutes a potential attack surface for indirect prompt injection. * Ingestion points: The skill ingests primary supplied evidence and current external research as defined in the Evidence discipline section of SKILL.md and references/skill-spec.json. * Boundary markers: The skill instructs the agent to explicitly tag material claims as fact, inference, assumption, or unknown, providing a logical separation between user input and model reasoning. * Capability inventory: The skill relies on external scripts (../../scripts/dynamic_router.py, ../../scripts/neural_router.py) and interacts with specialized agents (market-architect.md, marketing-skeptic.md). * Sanitization: The Evidence discipline policy acts as a logic-level boundary for processing external content.
[COMMAND_EXECUTION]: The skill instructions direct the model to utilize project-local Python scripts (../../scripts/dynamic_router.py and ../../scripts/neural_router.py) for building directed acyclic graphs (DAGs) and performing neural routing based on the request context. These are relative path references within the skill's expected host environment.

segmentation-strategy