The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data to build measurement architectures, creating a surface for indirect instructions.
Ingestion points: User-provided primary evidence is collected during the 'evidence' workflow step defined in references/skill-spec.json and governed by the Evidence Discipline in SKILL.md.
Boundary markers: The SKILL.md operating contract requires separating evidence from inference, and the references/decision-model.md mandates testing competing explanations, which acts as a logical check against malicious data, though no programmatic boundaries are defined.
Capability inventory: The skill utilizes several local scripts for routing (dynamic_router.py, neural_router.py) and invokes specialized agents (measurement-strategist, marketing-skeptic) and hooks (evidence-gate), providing a broad execution path for potentially injected instructions.
Sanitization: There is no evidence of automated sanitization, escaping, or filtering of user-provided text before processing.
[COMMAND_EXECUTION]: The skill directs the agent to execute local Python scripts to handle complex routing and dependency management.
Evidence: SKILL.md and references/skill-spec.json explicitly reference the use of ../../scripts/dynamic_router.py, ../../scripts/neural_router.py, and ../../scripts/skill_router.py to build bounded DAGs and select specialists. These scripts reside outside the skill's immediate directory but are integral to its operation.

marketing-measurement