一键导入
message-routing
Use when classifying user messages at process gates. Carried by orchestrator agents to determine how to handle user responses.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use when classifying user messages at process gates. Carried by orchestrator agents to determine how to handle user responses.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when creating a local agent for a pas-clief workspace. Walks the user through naming the agent, defining its role, listing its skills, and writing CLAUDE.md and CONTEXT.md from templates. The agent lands in `library/agents/<name>/`.
Use when drafting and publishing a newsletter issue. Walks research → draft → publish phases, using the writer agent. Example library process for pas-clief — copy or extend it for your own use.
Use when applying a workspace's drafting conventions to written output — voice, format, sentence rhythm, the things that make output feel like it belongs to this project. Reads the workspace's style notes if any and adjusts the draft accordingly.
Use when line-editing a draft for clarity and economy. Tightens sentences, removes filler, rewrites awkward constructions, ensures every sentence earns its keep. Apply this skill on the final pass before output.
Use when an agent needs to gather sources or background information from the web for a topic. Produces a concise sources file with URLs, summaries, and the most relevant excerpts.
Use when checking that a pas-clief workspace is current with the latest plugin conventions, or when migrating from an older version. Walks a checklist of structural and template changes and offers to apply each.
| name | message-routing |
| description | Use when classifying user messages at process gates. Carried by orchestrator agents to determine how to handle user responses. |
Classify user messages at gates to determine the correct response. Used by orchestrator agents during supervised mode when a gate pauses for user input.
Signals: "looks good", "continue", "approved", "go", "next", "yes", "lgtm", thumbs up, or any clear indication of satisfaction.
Behavior: Proceed to the next phase. No further action needed at this gate.
Signals: "this is too promotional", "the tone is wrong", "needs more sources", or any critique of the output quality.
Behavior:
.pas/workspace/{process}/{slug}/feedback/). Use the self-evaluation signal format:
Target: pointing to the skill or agent that produced the outputSignals: "why this angle?", "what sources did you use?", "can you explain...", or any request for information about the output.
Behavior: Answer the question using context from the phase output and agent work. Then re-present the gate. Do not proceed until the user gives an explicit approval or instruction.
Signals: "also include the SEC filing", "add a quote from...", "change the headline to...", or any directive to modify or extend the output.
Behavior: Incorporate the instruction into the current phase output. Route to the appropriate agent if needed. Re-present the gate after the change is made.
If the user's response does not clearly fit one classification:
A single message can contain multiple classifications. Handle them in order:
Example: "Why did you use that source? Also add the SEC filing. Otherwise looks good." = Question + Instruction + Approval. Answer the question, add the SEC filing, then proceed.