anomalyco / anomalyco/opencode
V2 beta: Transport error / hang when spawning multiple custom-agent subagents with large inline prompts
@rekram1-node is already working on this.
Since Aug 20, 2026.
- Dominant language
- TypeScript
- Stars
- 209k
- Forks
- 27.5k
- Avg merge
- 7h 2m
- Merged PRs (30d)
- 384
Description
Description
V2 beta: spawning a custom agent as a subagent multiple times (parallel or sequential) with a large inline prompt kills the session with a bare Error: Transport, or hangs indefinitely. The same workflow completes reliably in OpenCode 1.18.15, so this is a v2 regression.
Impact: multi-agent review workflows (spawn N reviewers, synthesize) are a core v2 use case and are unusable when the session dies after 1-2 subagent spawns.
Plugins
None (reproduced with zero plugins and zero MCP servers).
OpenCode version
opencode2 v0.0.0-beta-17639 (npm @opencode-ai/cli@beta). Works in opencode 1.18.15.
Steps to reproduce
- Config: one OpenAI-compatible provider (any local vLLM/Ollama endpoint), one custom agent with a system prompt and
mode: "all", subagent permission on the default agent:
{
"$schema": "https://opencode.ai/config.json",
"model": "local/my-model",
"default_agent": "build",
"providers": {
"local": {
"name": "local",
"package": "@opencode-ai/ai/providers/openai-compatible",
"settings": { "baseURL": "http://127.0.0.1:PORT/v1" },
"models": { "my-model": { "name": "My Model" } } // replace with a model your endpoint serves
}
},
"agents": {
"repro_reviewer": {
"description": "Custom reviewer agent",
"mode": "all",
"system": "You are a code reviewer. Review the provided diff for correctness, regressions, and missing tests.",
"permissions": [
{ "action": "shell", "resource": "*", "effect": "allow" },
{ "action": "edit", "resource": "*", "effect": "allow" }
]
},
"build": {
"permissions": [{ "action": "subagent", "resource": "*", "effect": "allow" }]
}
}
}
-
Run:
OPENCODE_CONFIG=opencode.json opencode2 run --standalone "Spawn repro_reviewer as a subagent THREE times IN PARALLEL. Each time pass this diff inline in the task prompt: <300-line diff> Wait for all three to complete. Report all three verdicts." -
Observed (8 runs):
- Parallel: 2 of 3 subagents complete, then
Error: Transport(2/3 runs); session hangs with no output (1/3 runs). - Sequential: completes ~4/5 runs, dies on the 2nd or 3rd spawn otherwise.
- Parallel: 2 of 3 subagents complete, then
Screenshot and/or share link
Server log shows the session ending with InterruptError: All fibers interrupted without error at Session.wait, http.status=499 (client connection closed), ~44s into the run. The CLI prints only Error: Transport with no further detail.
Operating System
macOS 26.6.2 (arm64)
Terminal
Ghostty (also reproduced headless via opencode2 run --standalone)
Additional notes (ruled out): the endpoint is not the cause (10+ concurrent direct HTTP requests all succeed ~0.5s); MCP is not the cause (zero servers configured); the custom system prompt alone is not the cause (small-task spawns of the same agent complete reliably). The trigger is repeated custom-agent subagent spawns combined with a large inline prompt.
Related: #13715, #35073, #36868 (subagent permission hangs), #38683 (transport retry), #38960 (subagent fan-out), #42725 (task tool swallows subagent errors).
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.