The agent says "We just need to wait for the result" -- and nothing else happens.

[AquaticAlpaca]

Overview

Problem: Agents quit midflow, but no errors or other problems are reported by BMAD, the IDE, or the model API.

What I expected: Agents to complete, or to provide a reason why continuing is impossible, or for the IDE or API to report an error.

Complication: I don't know where the fault lies: Are agents failing silently? Is the IDE eating errors? Is the model API failing to return a response?

Questions

• Am I using the agent wrong?
• Is there another level of verbose output that I can turn on?
• Is BMAD or Claude Code incompatible with my chosen model/API?

Configuration

• Linux
• BMAD-Method (v6.7.1) installed via npx
• Claude Code, running in a terminal inside VS Code (Codium, to be exact)
  • Model is openrouter/free (which automatically routes to one of several free models on OpenRouter)
  • /thinking is disabled
  • /effort is medium, but the problem also occurs on /effort high. Other /effort settings are untested
  • Verbose output is on

The Project

The problem occurs in multiple agents in a large project. To limit variables, I created a simple "Hello World" project.

Via BMAD, I created documentation for an application will create a static HTML doc to output "Hello, World" when run in a browser. I have the following completed documents in _bmad-ouput:

• prd.md
• project-context.md
• decision-log.md
• planning-artifacts/ux-design-specification.md
• planning-artifacts/epics.md
• planning-artifacts/implementation-readiness-report-.md
• implementation-artifacts/sprint-status.yaml

The next step is to create stories for the dev agent, so I executed /bmad-create-story in a new session.

The agent crunches for a while and returns this:

User didn't ask anything yet; they just invoked bmad-create-story skill which started workflow. Probably now they expect story creation. Need to run the skill. Use Skill tool with skill name bmad-create-story.

● Skill(bmad-create-story)
⎿ Successfully loaded skill

✻ Thinking…

● The story creation workflow has been started. I’ll let the skill run and will share the results once it finishes.

∴ Thinking…

We need to run skill, it will output result.

✻ Thinking…

✻ Baked for 1m 0s

And nothing else happens. The lack of additional output clashes with the agent's idea of what's happening ("I'll let the skill run and will share the results...").

To give the agent additional context, I re-ran the skill with a specific story: /bmad-create-story 1-1-implement-greeting-ui

It ran for a while and exited with this output:

● The “implement‑greeting‑UI” story is being created. I’ll let
the bmad‑create‑story skill finish its analysis and generate the story
document, then I’ll report back with the completed story file and next
steps.

∴ Thinking…

We need to run the skill with argument story. Already invoked, now need to let it process. Probably need to run with args? The skill already got args.
It will process. We just need to wait for result. No tool needed.

✻ Thinking…

✻ Crunched for 38s

And nothing else happens. In both cases, the agent seems to think another process or skill is running, but as far as I know only one skill is running at a time. Is there some parallelization I've overlooked?

If I keep re-running the skill in the same context, eventually it will produce a story. I don't expect deterministic results from an AI agent, but it seems reasonable that the framework or IDE or API would enforce a certain type of output each time, and I'm not seeing that happening.

What am I missing? Is this typical behavior, or is there a way to use the framework that produces a useful result each time an agent runs?

[zedshy]

i believe its multiple variables that is creating the issue