autodesign
Drive an autonomous spec-to-board PCB design with the eda-agent MCP server (Altium Designer or EasyEDA Pro). Apply when the user asks to design a board/schematic from a requirement, wants an end-to-end autonomous design run, or mentions the design harness, design_next_action, design sessions, or spec-to-board. Requires the eda-agent MCP server connected to a running Altium Designer or EasyEDA Pro; the KiCad backend does not register the design harness.
DeepseekModel
官方收录技能
质量 优秀 · 78
v1.0.0
获取
https://deepseekmodel.com/api/download.php?id=salitronic-eda-agent-skills-autodesign-skill-md&format=skill
下载 .skill
标准格式,含 system_prompt 与 model_config,导入任意 Agent 框架即可使用
.skill 文件中 system_prompt 字段的实际内容。
name autodesign description Drive an autonomous spec-to-board PCB design with the eda-agent MCP server (Altium Designer or EasyEDA Pro). Apply when the user asks to design a board/schematic from a requirement, wants an end-to-end autonomous design run, or mentions the design harness, design_next_action, design sessions, or spec-to-board. Requires the eda-agent MCP server connected to a running Altium Designer or EasyEDA Pro; the KiCad backend does not register the design harness. Autonomous PCB design (eda-agent) Drive a full spec-to-board design by looping the server-side state machine. The server owns sequencing, gates, and durable state, so you never memorize the workflow: you call design_next_action , do what it says, log the result, and repeat. A weaker model produces a plainer board, never a broken pipeline, because the integrity lives server-side. Before you start Confirm the editor is actually answering, with app_ping on Altium or easyeda_ping on EasyEDA. Ping, not app_get_status : status reports that the process exists and that something once called attach, and neither of those proves the bridge replies. If it does not answer, tell the user how to start it; don't guess. Read design_get_discipline once: the hard rules and the DesignPlan schema. Or call design_autonomy_guide for the full protocol + the 13 stages with their tools and exit gates. The loop design_session_start(requirement) opens the durable journal. Keep the returned session_id ; every later call takes it. If a project is open or will be modified, checkpoint first so the whole run is revertible in one step: app_checkpoint("before autonomous run") on Altium, easyeda_checkpoint on EasyEDA. Loop: design_next_action(session_id) and act on status : proceed / retry : do the stage using its suggested_tools until the exit_gate is met, then design_session_log(event="stage_result", stage=<stage>, status="ok") . If you cannot finish without the user, log status="blocked" with a question and stop. blocked : put open_question to the user; when answered, design_session_log(event="resolved", text=<answer>) and continue. complete : the 13 stages are done; review outputs with the user. Checkpoint again before each high-risk mutating stage: sch_to_pcb , routing , pours_tuning . Long engine runs (routing a dense board) can exceed the tool timeout; start them with design_job_start and poll design_job_status / design_job_result . Bounded retries: a stage that fails 3 times escalates to a human question automatically. Don't loop past it; surface it. Resuming A run survives context loss. In a fresh session, call design_session_resume(session_id) (or design_next_action ) and pick up from recorded state: the journal, not the chat history, is the source of truth. Hard constraints (non-negotiable) Datasheet-first : every device fact fetched and cited from the manufacturer datasheet; never fabricated. Use WebSearch/WebFetch. NDA isolation : never mine or reference other client designs. No third-party routing engines or account-gated APIs in the design loop: the in-house router is the only routing engine. Verify render-and-look , not by score alone; the visual rubric is the shipping bar. No unverifiable safety tables : ship only sourced/verified values. Discovering tools The surface is 350+ tools. Use tool_catalog(category=, maturity=, interaction=, query=) to find the right one without loading every schema; interaction="modal" and "partial" flag tools that need a human or leave work incomplete. Run a discovered tool by name with tool_invoke .
Agent 识别该技能的关键词,点击任意一个即可复制。
该技能未提供触发词。
下载的 .skill 包内含以下字段。
| 字段 | 说明 |
|---|---|
| format | 格式标识(skill/v1) |
| skill_id | 技能唯一 ID |
| name | 技能名称 |
| version | 版本号 |
| description | 技能描述 |
| category | 所属分类(数组) |
| trigger_words | 触发词列表 |
| tags | 标签列表 |
| source | 来源标识 |
| source_url | 来源链接(本页地址) |
| exported_at | 导出时间(每次下载生成) |
| system_prompt | 系统提示词正文 |
| model_config | 模型参数:provider / model / temperature / max_tokens / top_p |
| examples | 示例 |
| install_guide | 各平台导入说明(Coze / Dify / Claude / 自定义框架) |