The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes a behavior specification file (docs/product/behavior-spec.md) that serves as the primary driver for autonomous activities, creating a surface for indirect prompt injection where malicious instructions in the document could influence agent behavior.
Ingestion points: The agent reads docs/product/behavior-spec.md and .agent-atelier/work-items.json to determine task goals.
Boundary markers: While an information barrier is mentioned between implementation and validation phases, no explicit delimiters or instruction-ignore markers are defined for the input specification itself.
Capability inventory: The skill and its spawned teammates utilize Bash, Write, and Agent (for spawning additional sub-agents) as specified in SKILL.md.
Sanitization: No sanitization or validation of the natural language instructions within the behavior specification is documented.

run