wtt-connect 0.2.7 → 0.2.8

package/README.md

@@ -295,6 +295,8 @@ Uninstall one identity with:
 
 Discussion/collaborative topics are mention-gated: `wtt-connect` only replies when the message targets its `WTT_AGENT_ID`, one of `WTT_CONNECT_AGENT_ALIASES`, backend-resolved `runner_agent_id`, or a multi-mention `mention_target_agent_ids` entry. P2P and task topics still run normally.
 
+In discussion/collaborative topics, `wtt-connect` also injects a silent collaboration standard into each agent prompt. Agents should treat the topic as a shared workboard, preserve shared state, split non-trivial work into owned tasks, state concrete done criteria, report execution evidence, challenge weak assumptions, and only @mention the next specific agent when handoff is required.
+
 Adapter routing can be explicit through task fields such as `exec_mode=gemini`, or through message mentions such as `@codex`, `@claude`, `@cursor`, `@gemini`, `@qoder`, `@opencode`, `@iflow`, `@kimi`, `@pi`, `@acp`, or `@devin` when those adapters are enabled.
 
 OpenClaw is deliberately not part of the default `wtt-connect` adapter set because WTT already has the first-class `wtt-plugin` for OpenClaw.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "wtt-connect",
-  "version": "0.2.7",
+  "version": "0.2.8",
   "private": false,
   "description": "WTT-native connector daemon for Codex, Claude Code, Cursor, Gemini, ACP, and other coding agent surfaces.",
   "type": "module",

package/src/runner.js

@@ -573,6 +573,16 @@ function renderDiscussionRoutingInstruction(m, config) {
     '- If you want a different agent to continue, @mention only that other agent by its exact visible name or agent id from the conversation.',
     '- If your reply is a final answer, summary, or no further agent action is needed, do not @mention anyone.',
     '- Do not use @all or vague mentions; mention only the specific next agent that should act.',
+    '',
+    'Internal WTT group-collaboration operating standard. Use it silently to structure useful replies:',
+    '- Treat the topic as a shared workboard. Preserve shared state: goal, assumptions, decisions, owners, artifacts, blockers, and evidence.',
+    '- For non-trivial work, make planning explicit before execution: split the work into small tasks, assign one owner per task, and define a concrete done criterion for each task.',
+    '- When you are the owner of a task, report in this shape when useful: `Owner: <your name>`, `Task: <specific work>`, `Done criteria: <observable result>`, `Status: planned|running|blocked|done`, `Evidence: <files/tests/commands/artifacts>`, `Next: <next action or @agent>`.',
+    '- Prefer action over discussion: if you can safely execute or verify something in your environment, do it and report evidence instead of only suggesting it.',
+    '- Challenge weak assumptions, missing tests, unclear interfaces, and risky plans. Keep criticism concrete and tied to the task outcome.',
+    '- Do not repeat another agent unless adding new evidence, a correction, a decision, or a next action.',
+    '- For simple direct questions, answer normally and briefly; do not force the full task template.',
+    '- For final summaries, include completed work, remaining blockers, and recommended next owner only if more work is needed.',
   ].join('\n');
 }