openclaw - 💡(How to fix) Fix Feature request: allow hooks to mutate bodyForAgent on message:preprocessed [1 participants]

Feature Request

Summary

Allow hook handlers listening to message:preprocessed to mutate context.bodyForAgent so that the modified content is what the agent actually receives.

Motivation

Currently, hooks can read context.bodyForAgent on message:preprocessed, but any mutations to the field are silently ignored — the agent always receives the original value. This makes it impossible to build per-message context injection hooks.

A concrete use case: a semantic memory recall hook that, for every inbound message, runs a vector search against past conversation history and appends the top relevant snippets to bodyForAgent. This would give the agent automatic, query-specific context without needing to manually call memory_search on every turn.

Current Behavior

Hook handler mutates context.bodyForAgent:

const handler = async (event) => {
  if (event.type !== 'message' || event.action !== 'preprocessed') return;
  const extra = await semanticSearch(event.context.bodyForAgent);
  // This mutation has no effect — agent sees the original value
  event.context.bodyForAgent += `\n\n---\n${extra}\n---`;
};

The mutation does not reach the agent.

Requested Behavior

Mutations to context.bodyForAgent in a message:preprocessed hook handler should be respected and passed to the agent as the final message body.

This mirrors the existing pattern for agent:bootstrap, where hooks can mutate context.bootstrapFiles and the changes are applied.

Workarounds Considered

• agent:bootstrap mutation: fires once per session, not per-message — can't be query-driven
• Manual memory_search tool calls: works but requires the agent to remember to do it; not automatic

Impact

This would unlock a whole class of dynamic context injection hooks (memory recall, RAG pipelines, live data enrichment) that are currently not possible.

Filed by Stella (OpenClaw pa agent) on behalf of aruizrab