dsh-session-mindmap

Usage

Ask the model:

把这个会话整理成脑图

or call the tool yourself:

session_mindmap
  sessionId?  "last" or a session id; defaults to the current session
  kinds?      CSV of topic,conclusion,decision,todo,question,file
  focus?      only summarise this topic
  language?   zh or en; defaults to the plugin config
  force?      ignore the cache and call the model again

or use the command, which never involves the model in deciding anything:

/mindmap                       # current session
/mindmap last                  # most recent session
/mindmap session-abc --reveal  # a specific session, then select it in the file manager
/mindmap --no-open             # write the file quietly
/mindmap --focus=发布方案       # only one topic
/mindmap --kinds=topic,file    # include the files that were touched
/mindmap --lang=en             # one English map, without touching the config
/mindmap list                  # recent sessions with their ids and titles

The command is the intended entry point: it runs even when you do not want to spend a model turn on the request itself.

Configuration

Set these in the plugin row (profile cordis.patch.yml) — a patch replaces the whole config object, so restate every key you want to keep:

- id: session-mindmap
  name: 'dsh-session-mindmap'
  config:
    language: en
    maxNodes: 120

Key	Default	Meaning
provider / model	empty	Empty follows the agent's default model. Point these at a cheap model if you generate these often.
kinds	topic, conclusion, decision, todo, question	Node kinds to extract. Add file to include the files a session touched.
language	zh	Node language: zh or en (the language tool argument overrides it for one run).
maxInputTokens	24000	Transcript budget for one model call.
maxBlocks	8	Hard cap on map calls for a long session, plus one merge call.
maxNodes / maxDepth	80 / 4	Clamps applied to whatever the model returns.
maxOutputTokens / temperature	8000 / 0.2	Per-call generation settings. A node costs roughly (label + detail) characters, so this cap has to hold the whole answer — see Troubleshooting.
llmTimeoutMs	180000	Per-call timeout.
outputDir	.dsh/mindmap	Relative to the session workspace; absolute paths are accepted.
cache	true	Reuse the last result while the session has not grown.
openAfterBuild	true	Open the artifact after a human runs /mindmap. Ignored for model-invoked calls, which leave a deliverable card instead.