The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes complex architectural decisions and design trade-offs provided by users, which are then distributed to multiple sub-agents representing different perspectives. This creates a surface for indirect prompt injection if the user-provided context contains instructions designed to bias or override the sub-agents' logic.
Ingestion points: User-provided inputs are ingested via the 'Decision Framing Template' (Decision, Context, Constraints, Evaluation Criteria) as described in SKILL.md.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' prompts when interpolating user context into the perspective profiles.
Capability inventory: The workflow involves spawning subprocesses for each perspective, using a 'builder agent' to implement the chosen approach (implying file system or environment modifications), and persisting data using store_discovery() from the amplihack.memory.discoveries module.
Sanitization: There is no mention of sanitization, escaping, or validation of the input data before it is passed to the sub-agents or the builder agent.

debate-workflow