openclaw - 💡(How to fix) Fix [Regression] Slack reply context not passed to model in v2026.5.4

Problem

Starting in v2026.5.4, the assistant loses context of which message the user is replying to in Slack. When a user replies to a specific message, the assistant responds as if it has no idea what the user is referring to.

Impact

Critical usability regression - the assistant cannot understand conversational context when users use Slack's native reply feature.

Root Cause

Commit a7851f2150 ("fix: keep runtime context out of model prompt") removed the code that queues runtime context for the next turn:

// This code was REMOVED in a7851f2150:
await queueRuntimeContextForNextTurn({
  session: activeSession,
  runtimeContext: runtimeContextForModel,
});

This runtime context includes the reply_to_id metadata that tells the model which message the user is replying to.

Related Commits

• a7851f2150 - "fix: keep runtime context out of model prompt" (May 6, 2026)
• 6dae3c273d - "fix: keep runtime prompt context out of system prompt (#77521)" (May 4, 2026)

The intent was to improve prompt caching by keeping runtime context out of the system prompt, but the implementation went too far and removed reply context entirely from the model's view.

Version Info

• Broken: v2026.5.4 (current)
• Working: v2026.4.26 (last known good)
• Platform: Slack DM conversations

Steps to Reproduce

1. Send a message in a Slack DM with the bot
2. Use Slack's reply feature to reply to that specific message
3. Observe that the bot's response shows no awareness of what message was replied to

Expected Behavior

The bot should receive the reply context and understand which prior message the user is responding to.

Suggested Fix

Restore the queueRuntimeContextForNextTurn call for reply metadata while keeping the prompt cache optimization for other runtime context. Reply context is conversation-critical and should be included in the model's prompt even if it breaks cache reuse.