Feature #1 is indefinitely in-progress

[dineshkp]

I've been test-driving auto-forge for about two hours now — really creative approach. Kudos and thanks for bringing it to life!

Issue: Feature #1 appears to be stuck in a loop. It is marked as done, yet an agent is always actively working on it at any given point in time.

Steps to reproduce:

1. Run auto-forge with the default configuration
2. Observe agent activity on Feature fix: auto-activate venv in start_ui.sh #1 over time

Expected behavior: Once a feature is marked as done, no agent should continue working on it.

Actual behavior: An agent is perpetually assigned to Feature 1 despite its completed status.

Additional context: I asked the built-in assistant to diagnose the issue. Here's its output:

Hope this is sufficient to help look into the issue further. Happy to provide any additional details if needed.

[pelzi68]

I assume this is the same issue that I have... I asked the Project Assistant to write a bug report on the repeating issues that I am having.

Coding agents not automatically picking up next feature after completing work

Description

Agents that complete a feature are not being freed/recycled to pick up the next ready feature. They appear to hang in Mission Control and on Done feature even though:

• The feature is marked as passes: true
• The feature is marked as in_progress: false
• There are ready features available in the queue

This has happened repeatedly (e.g., features 55, 56, 57 all completed but agents did not proceed to the next work item). It requires manual intervention to restart agents each time (stop -> start), which defeats the purpose of autonomous operation.

Expected behavior

After an agent marks a feature as passing, it should automatically claim and begin work on the next ready feature.

Actual behavior

Agents appear to stop/hang after completing a feature. Mission Control shows them as still occupied despite the feature being fully resolved.

Impact

All forward progress halts until agents are manually restarted. Currently 18 features are ready but 0 agents are working.

Steps to reproduce

1. Launch a coding agent via Mission Control
2. Let it complete a feature (mark as passing)
3. Observe that the agent does not pick up the next ready feature
4. Agent remains idle in Mission Control

Environment

• Autoforge with Claude coding agents
• Feature backlog managed via MCP feature server

[dineshkp]

This definitely sounds related. Interesting that we're seeing two sides of the same coin:

• My case: Agent stays stuck actively reworking a feature that's already marked as done.
• Your case: Agents go idle after completion and don't pick up the next one.

Both seem to stem from the same area. The assistant's analysis in my original post traced it to a SQLAlchemy session cache invalidation issue in feature_mcp.py — agents running as separate MCP processes read stale data because session.expire_all() isn't called in the query functions. The orchestrator handles this correctly, but the MCP server functions don't.

Your idle-agent behavior could well be the flip side of this — if the agent's session doesn't see updated feature states, it may not recognize there's new work available.

Would be curious to know if you're also running multiple agents in parallel.

[pelzi68]

@dineshkp,

yes three agents in parallel and one agent working on max two features per session.