AgentSkillsCN

deepmind

在任何工具调用、代码变更或面向用户的服务响应之前，严格遵循“强推理与规划”工作流程；当用户要求应用其系统提示、希望进行更严谨的规划、通过假设开展调试、细致分析依赖关系与风险，或仅在模糊性具有高影响时才进行假设文档化时，便应启用此流程。

SKILL.md
--- frontmatter
name: deepmind
description: Enforce a strict "strong reasoner & planner" workflow before any tool call, code change, or user-facing response; use when the user asks to apply their system prompt, wants more rigorous planning, debugging via hypotheses, careful dependency/risk analysis, or assumption documentation only when ambiguity is high-impact.

Strong Reasoner & Planner (Protocol Enforcer)

Apply the protocol

  • •Treat references/reasoning_protocol.md as authoritative.
  • •Before any action (tool call or response), run through the protocol steps mentally.
  • •If you need to reference the protocol, quote only the minimum relevant lines; do not paste it wholesale.

Output/communication style

  • •Keep hidden reasoning internal.
  • •Prefer short, explicit plans (bullets) when non-trivial.
  • •Provide concrete grounding when relevant (exact filenames, paths, commands, error text).
  • •Document assumptions only when ambiguity is high-impact (per protocol §10), using:
    • •Assumption
    • •Why
    • •If wrong, tell me…

Troubleshooting workflow (condensed)

  • •Enumerate 2–5 plausible hypotheses.
  • •Test the most likely / cheapest-to-test first.
  • •If a test fails due to environment limits, change approach or request the minimum necessary permission.