The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a workflow for processing untrusted user data, such as code diffs, architectural plans, and bug descriptions, which are then briefed to subagents. This ingestion process is an indirect prompt injection surface.\n
Ingestion points: The 'Package the Dilemma' section in SKILL.md instructs the agent to intake user-provided context, objectives, constraints, and options.\n
Boundary markers: The skill includes explicit instructions to 'never dump the raw chat history' and to 'cleanly isolate and distill' the context, providing a structural boundary for the data.\n
Capability inventory: The instructions reference the invoke_subagent tool and background agent execution as the primary mechanisms for processing the distilled data.\n
Sanitization: The mandatory 'Anti-Bias Briefing' and distillation process function as a sanitization layer by filtering extraneous or potentially malicious instructions before they reach the subagents.

consult