Agent: CLI log stuck at generateSpec during loop execution

[federiconeri]

Summary

When the agent runs a loop via runLoop, the CLI agent log shows [tool:done] generateSpec → true and then shows no further updates until the entire loop completes, leaving the user with no visibility into loop progress.

Problem / Context

During the second agent test, after generateSpec completed:

• Agent log displayed: [tool:done] generateSpec → true
• Then silence for ~12 minutes while the loop ran 3 sessions
• Finally: [tool:done] runLoop → done

The user had to check the monitor command separately to see any progress. The agent's own stream provided zero intermediate feedback during the most time-consuming phase.

Root Cause

The runLoop tool in src/agent/tools/execution.ts spawns feature-loop.sh as a subprocess with stdio: 'pipe' but doesn't stream the subprocess output back to the agent's event stream. The tool only returns the final result.

Proposed Solution

1. Stream subprocess output: Pipe feature-loop.sh stdout to the agent's event emitter as progress events
2. Periodic status updates: Poll the loop's status files (/tmp/ralph-loop-*.status, .phases) and emit structured progress events
3. Phase transition events: Emit events when the loop transitions between phases (planning → implementation → verification → PR)

Files to Modify

File	Changes
src/agent/tools/execution.ts	Stream loop subprocess output as progress events
src/commands/agent.ts	Handle and display progress events from tools

Acceptance Criteria

• Agent log shows phase transitions during loop execution
• User can see loop progress without checking monitor separately
• At minimum: "Loop phase: planning", "Loop phase: verification", etc.