fix(job_output): add timeout to prevent indefinite blocking on wait=true

[kkmkoi]

Problem

job_output(shell_id="X", wait=true) blocks forever when used on a long-running process (e.g. a web server, dev server, or any process that never exits).

This is a particularly nasty bug because:

• Short commands (build, test, git) complete instantly → never trigger it
• Long-running processes (start server, run watcher) → exactly the scenario where users ask an AI coding assistant to "run my project" → deadlock

The root cause: WaitContext only returns when:

1. The process exits — server never exits → never fires
2. The context is cancelled — no timeout is set → never fires

Fix

Added a timeout_seconds parameter to JobOutputParams (default: 60, max: 600). When wait=true, the WaitContext call is now wrapped with context.WithTimeout, so the tool always returns within the configured timeout.

Changes

File	Change
job_output.go	Added TimeoutSeconds field + context.WithTimeout wrapper
job_output.md	Updated description to mention timeout
bash.go	Added MaxAutoBackgroundSeconds to template data
bash.md.tpl	Added warning: NEVER use wait=true on server processes

Example

# Before: blocks forever on a running server
job_output(shell_id="022", wait=true)  → never returns

# After: returns within timeout
job_output(shell_id="022", wait=true)                → returns after 60s
job_output(shell_id="022", wait=true, timeout_seconds=300) → 5min

[charmcli]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[kkmkoi]

recheck

[kkmkoi]

I have read the Contributor License Agreement (CLA) and hereby sign the CLA.

[kkmkoi]

recheck

[kkmkoi]

I encountered this bug while working on a real project — a customer service AI assistant for a beauty e-commerce competition (欧莱雅 Beauty Techathon).

The workflow was:

1. I asked Crush to build and run my Go backend server
2. The build completed quickly ✅
3. The server started and was moved to background (shell_id="022") ✅
4. Crush called job_output(shell_id="022", wait=true) to check if it started successfully
5. The server was running fine — but job_output never returned because a server process never exits
6. I waited 10+ minutes before realizing it was deadlocked 😅

The fix with timeout_seconds would have saved me those 10 minutes — the tool would have returned with the server's startup logs after 60 seconds, and Crush could have confirmed it was running from the output.

This is a great project for learning and contributing to open source. Thanks for maintaining Crush! 🙏