Bug: On Windows, when adding an Agent from github, Agent builds, but fails to run

[KenOcheltree]

Pre-requisities

• [ X] I am using the newest version of the platform (agentstack version shows that CLI and platform are the same version, and there is no newer version available)

Describe the bug
Running Agentstack on WIndows, give agentstack add command with github location of agent, the build process completes, but an error occurs when the agent run is attempted.

This identical behavior was seen on two different Windows PCs. Also, the same command works fine on a MAC system.

To Reproduce
Steps to reproduce the behavior:

1. Start Agentstack running with Google gemini as the LLM and using default models.
2. Issue the command;
   agentstack add https://github.com/sandijean90/AgentStack-HealthcareAgent@release-0.0.16#path=/provider_agent
3. The agent will build, but fails to start

Expected behavior
The agent should start running and should show up in agent list

Logs / Screenshots / Code snippets
On agentstack add, the following output and error is received after the successful build:

[kaniko-build]: INFO[0014] Taking snapshot of full filesystem...
[kaniko-build]: INFO[0029] ENV PRODUCTION_MODE=True     PATH="/app/.venv/bin:$PATH"     HOME=/tmp
[kaniko-build]: INFO[0029] CMD ["uv", "run", "--no-sync", "server"]
[kaniko-build]: INFO[0031] Skipping push to container registry due to --no-push flag
[crane-push]: Starting log stream...
[crane-push]: 2025/12/31 18:05:30 existing blob: sha256:6818054bc0c13af208f27120ebfbca4dfc23254c30e9df6c5447877f0cd4dba8
[crane-push]: 2025/12/31 18:05:30 existing blob: sha256:5527b5a510395b418301526e174ae1df6285f013b551009924a50936600541bb
[crane-push]: 2025/12/31 18:05:30 existing blob: sha256:02d7611c4eae219af91448a4720bdba036575d3bc0356cfe12774af85daa6aff
[crane-push]: 2025/12/31 18:05:30 existing blob: sha256:f1a1869fe2bff975df64319af2b9ed2bb1bf2ed6be99b87d21c9aa94d45a72b4
[crane-push]: 2025/12/31 18:05:30 pushed blob: sha256:5db5aa2b4e5aba199d1b8172f73a50e628024610bbca53b7d3557a969dfd8f2f
[crane-push]: 2025/12/31 18:05:30 pushed blob: sha256:b4377757205fdcc3e421c2ff7adec90b5047e562ab0e603ed5dfe744dcddedfa
[crane-push]: 2025/12/31 18:05:30 pushed blob: sha256:b946ba2eb7d0501f2e3516b3f004803954cf3b5f4ccd4f9d2adb1f65f86cbbf7
[crane-push]: 2025/12/31 18:05:30 pushed blob: sha256:26aa120ae788de7fc65206fdd5472743416846dd5340e714c9b1d5276b784c2b
[crane-push]: 2025/12/31 18:05:30
agentstack-registry-svc.default:5001/sandijean90/agentstack-healthcareagent/provider_agent:a843e4b124d5a3c36240e64a803e5
e04978d8805: digest: sha256:b7a741158c3aba1b98f4065cede06e4848ab26094d73d7ef9e125eba90277b81 size: 1412
[crane-push]:
agentstack-registry-svc.default:5001/sandijean90/agentstack-healthcareagent/provider_agent@sha256:b7a741158c3aba1b98f406
5cede06e4848ab26094d73d7ef9e125eba90277b81
Job failed.

------- Error -------

💥 ValidationError: 1 validation error for ProcessLogMessage
message
  Input should be a valid string [type=string_type, input_value=[('ServerError', 'contain...tart: PodInitializing')],
input_type=list]
    For further information visit https://errors.pydantic.dev/2.12/v/string_type

If agent is not working, please send output of the following commands:

agentstack list
📝 INFO: Listing agents on http://localhost:8333
SHORT ID  NAME                    STATE          LOCATION           INFO
2158c059  ChatV2                  ○ not started  agents/chat:0.4.3  <none>
e8d6f79b  RAG                     ○ not started  agents/rag:0.4.3   <none>
a9b589ae  Single-turn Form Agent  ○ not started  agents/form:0.4.3  <none>

If applicable, add screenshots or code snippets to help explain your problem.

Set-up:

• Model provider: Google gemini using default models

Additional context
Add any other context about the problem here.

[dosubot]

The root cause is a bug in how agent startup errors are logged: when an exception occurs during agent startup, the code tries to log the error by passing the output of extract_messages(ex) directly to the ProcessLogMessage model. On Windows, this function often returns a list of (type, message) tuples (especially when asyncio raises an ExceptionGroup), but the model expects a string. This mismatch triggers the Pydantic ValidationError you saw, which prevents the agent from starting and showing up in the list. This is less likely to happen on Mac because of differences in how exceptions are grouped and surfaced by the event loop.

To fix this, the code should convert the list of exception messages into a single string before passing it to ProcessLogMessage. For example, in the relevant error handler, replace:

ProcessLogMessage(stream=ProcessLogType.STDERR, message=extract_messages(ex), error=True)

with something like:

ProcessLogMessage(
    stream=ProcessLogType.STDERR,
    message="; ".join(f"{t}: {m}" for t, m in extract_messages(ex)),
    error=True
)

This will ensure the message field is always a string, preventing the validation error and allowing agent startup logs to be processed correctly on Windows. There is no evidence that the Docker image or runtime environment is the source of the bug—it's a cross-platform logging issue in the Agentstack codebase itself.
See supporting code references

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}