Feature: Allow sending follow-up messages during active Claude task

[vckobi]

Feature Request

Problem

When Claude is actively working on a task (reading files, editing code, running commands), the user cannot send additional context, corrections, or notes until the current operation completes.

Claude Code CLI natively supports this — you can type messages while Claude is working and it picks them up. However, the Telegram bot currently blocks input during active tasks, requiring fully sequential interaction.

Proposed Solution

Allow users to send messages while Claude is processing a request. These messages could either:

1. Queue & append — Collect follow-up messages and feed them to Claude as additional context when the current turn ends (simplest approach)
2. Live inject — Pass the message to the active Claude session in real-time, similar to how the CLI works (ideal but more complex)
3. Hybrid — Let short messages queue, but support a prefix like ! or /note to inject immediately

Use Cases

• User realizes they forgot to mention a constraint → sends "also make sure to handle null values"
• User wants to redirect Claude mid-task → "actually, use TypeScript instead"
• User wants to add context → "the config file is at /etc/app/config.yaml"
• User spots Claude going in wrong direction → "stop, that's the wrong file"

Related PRs

• feat: add /stop command to cancel running Claude calls #116 (/stop command) — covers cancellation but not appending context
• feat: add inline Stop button to cancel running Claude requests with native Telegram UI #122 (inline Stop button) — same, cancel only
• feat: stream partial responses via Telegram sendMessageDraft API #123 (streaming responses) — shows progress but no input during task
• Enable concurrent processing of messages across topics #102 (concurrent processing across topics) — parallel topics, not mid-task input

Environment

Using claude-code-telegram v1.3.0 inside a Docker container (linuxserver/code-server) with Claude Code CLI.

[xiaotonng]

Standard stdin blocking makes this tricky. In pikiclaw, we solved this by decoupling the IM event queue from the stream loop via an async watchdog, allowing interruptions or follow-ups via Telegram mid-generation. Feel free to reference our approach or test it via npx pikiclaw@latest.