The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests conversational user intent to generate a plan that dictates which specialized skills are loaded and executed. This creates a surface where user-provided instructions could influence the selection or sequence of operations. The skill mitigates this through a structured brainstorming phase, mandatory use of reference contracts, and a manual approval step where the user must review the plan before it is written to disk or executed.
Ingestion points: User goals, requests, and feedback provided during the conversational brainstorming phase as defined in Phase 1 and Phase 3 of SKILL.md.
Boundary markers: The skill explicitly references input-output-contracts.md and skill-routing-constraints.md to validate task ordering and prerequisites. It also requires the user to choose between specific workflows (evaluate-first vs direct fine-tuning) before a plan is generated.
Capability inventory: The skill has the capability to write to PLAN.md and load external skill files (SKILL.md) for execution as described in the Phase 2 and Execution sections of SKILL.md.
Sanitization: There is no explicit text sanitization mentioned; however, the design incorporates a human-in-the-loop review of the proposed plan before execution.

planning