Ralph not starting on existing project

[Calfredop]

Hello,
First of all thanks for your work, the potential here looks enormous. I am trying to integrate ralph into an existing codebase, so I have provided a an md file with project scope and spect, let the ralph-import do its job, then proceeded to refine generated files, creating a specs/requirements.md with current project stack(s) and rearranging tasks in @fix_plan.md.
I then tried to start ralph with monitoring, but the loop interrupts on first Claude Code Call. After I call the command I get this log:
[2026-01-15 22:58:44] [SUCCESS] 🚀 Ralph loop starting with Claude Code [2026-01-15 22:58:44] [INFO] Max calls per hour: 100 [2026-01-15 22:58:44] [INFO] Logs: logs/ | Docs: docs/generated/ | Status: status.json [2026-01-15 22:58:44] [INFO] Starting main loop... [2026-01-15 22:58:44] [INFO] DEBUG: About to enter while loop, loop_count=0 [2026-01-15 22:58:44] [INFO] DEBUG: Successfully incremented loop_count to 1 [2026-01-15 22:58:44] [INFO] Loop #1 - calling init_call_tracking... [2026-01-15 22:58:44] [INFO] DEBUG: Entered init_call_tracking... [2026-01-15 22:58:44] [INFO] DEBUG: Completed init_call_tracking successfully [2026-01-15 22:58:44] [LOOP] === Starting Loop #1 === [2026-01-15 22:58:44] [INFO] DEBUG: Checking exit conditions... [2026-01-15 22:58:44] [INFO] DEBUG: Exit signals content: {"test_only_loops": [], "done_signals": [], "completion_indicators": []} [2026-01-15 22:58:44] [INFO] DEBUG: Exit counts - test_loops:0, done_signals:0, completion:0 [2026-01-15 22:58:44] [INFO] DEBUG: @fix_plan.md check - total_items:49, completed_items:33 [2026-01-15 22:58:44] [INFO] DEBUG: No exit conditions met, continuing loop [2026-01-15 22:58:44] [LOOP] Executing Claude Code (Call 1/100) [2026-01-15 22:58:44] [INFO] ⏳ Starting Claude Code execution... (timeout: 15m) [2026-01-15 22:58:44] [INFO] Using modern CLI mode (JSON output)

Then the output hangs for a few seconds on the last line, just to exit back to terminal. Progress and status files are set to running status:
{ "status": "executing", "indicator": "⠋", "elapsed_seconds": 10, "last_output": "", "timestamp": "2026-01-15 22:58:44" }
{ "timestamp": "2026-01-15T21:58:44+00:00", "loop_count": 1, "calls_made_this_hour": 0, "max_calls_per_hour": 100, "last_action": "executing", "status": "running", "exit_reason": "", "next_reset": "23:58:44" }

[coderabbitai]

📝 CodeRabbit Plan Mode

Generate an implementation plan and prompts that you can use with your favorite coding agent.

• Create Plan

Examples

• Example 1
• Example 2

---

🔗 Related PRs

#90 - fix(loop): respect Claude's EXIT_SIGNAL when checking completion indicators [merged]

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[mcapodici]

Same here. In my case it is doing something, as it got me rate limited. I asked another claude instance to confirm another claude was running. It is just not clear that progress was being made.

[elvyyy]

Same issue

[xujiangze]

Same issue

[falconmick]

same

[ShayneP]

same

[0xAnton1]

I have a working solution for your y'all @falconmick @ShayneP @xujiangze

Use any LLM to create the spec and list

The original metaprompt in ralph is detailed enough for any LLM to create the required docs for ralph to run properly. Here is how you do it.

Step 1 ; On cursor or equivalent:

• Ask the LLM to review your repo and code and give it an instruciton to summarize the spec etc, in a new file OVERVIEW.md
• check if it works for you, and add the futures steps or features you wish to add

Step 2 ; Add the ralph setup to your repo , needs to be a git repo already.

• in your repo, setup the ralph project ---> ralph-setup "your-ralph-project, it will create the initial file but empty folder for spec and to do.
• the main files you need to worry about are here:
  `
  File Structure
• specs/: Project specifications and requirements
• src/: Source code implementation
• examples/: Example usage and test cases
• @fix_plan.md: Prioritized TODO list
• @AGENT.md: Project build and run instructions

Step 3: prompt something like this:

• in your LLM chat on cursor ask, " Tell me how we can initiate @"your-ralph-project"/PROMPT.md process by taking our existing @overview.md and see what is left to do, to create hte task list etc ? I am not sure how to use the tool from https://github.com/frankbria/ralph-claude-code.git to effectively create steps. "
• this will five you a fix_plan.md, the spec.md and a customized version of @agent.md, for your specific conda, venv etc. So the LLM can use the bash / terminal properly.
• the LLM will be able to give you a very detail list fitting the template for ralph you can review and edit, move it to the folders, or let it write directly.

Step 4: Run Ralph and see its progress

• ralph --monitor (this initiate the workflow)
• this should do it

Enjoy

[frankbria]

This issue may be related to #121 - the need for a ralph-init command to properly initialize Ralph in existing projects.

Based on the logs, Ralph starts normally but hangs at 'Using modern CLI mode (JSON output)' and then exits silently. This suggests the Claude Code CLI invocation may be failing.

Questions to help diagnose:

1. Is Claude Code CLI working standalone? Try: npx @anthropic/claude-code --version
2. Is your existing project a git repository?
3. Were the .ralph/ files copied manually or created via ralph-setup?
4. What's in .ralph/PROMPT.md? (First 20 lines would help)