[FEATURE] FireHose monitoring. Add the abilty for other processes watch all communication with the AI.

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

This is a feature distinct from, but might be a solution to many of the problems have with Claude showing insufficient thinking. There are a lot of Issues regarding the loss of functionality through increased conciseness in the output #8477 #8586 #33163 #8371.

Much of the debate seems to be about where the appropriate place for this information should be. As Claude works for longer and uses more tools, presenting the information inline has the potential to swamp other data.

Proposed Solution

The solution I propose is a Firehose. This could be implemented by any manner of interprocess communication, the most simplest would be to have Claude Code listen on a socket and give a copy of everything to all connections that attach to the socket.

Possibly even simpler would be just having a FIREHOSE.md (or .txt or just plain FIREHOSE) file that would receive everything, but at the cost of wasting space if nobody every looks at it. Perhaps as an option this could be provided, but I would lean towards some form of interprocess communication, because this would be less about what happened in the past but rather what is going on now.

The Firehose should emit the communication with the AI soon as it appears (buffering batches of characters for efficiency notwithstanding). The point is it should be up to date to about a second or so.

This output would be possibly extremely high volume, but that fact is indicated by the name Firehose. What users do with that output is up to them. Many would just want to make a pretty printed scrolling window just to give a sense of what is happening right now. Some might want to scan for keywords to alert them to certain actions, Some might pass to a local LLM for a "what's going on right now" summariser.

It's an easy to implement Feature that enables users to develop their own inspection interface to use in conjunction with Claude Code.

Priority

High - Significant impact on productivity

Feature Category

Interactive mode (TUI)

Use Case Example

A user might write a viewer window that sits across the top of the screen showing the last few lines of communication so that they can get an idea of what is happening within the internal monologue of the AI.

A more advanced inspector may monitor the Firehose to perform a separate analysis of what tools were used and evaluate the level of success the model had in using each tool.