subagent: tool remains advertised at the maximum session depth #53356

Activity

1. github-actions

   assigned

   rekram1-node

   on Oct 5, 2026
2. github-actions

   added

   needs:complianceThis means the issue will auto-close after 2 hours.This means the issue will auto-close after 2 hours.

   and removed

   needs:complianceThis means the issue will auto-close after 2 hours.This means the issue will auto-close after 2 hours.

   on Oct 5, 2026
3. github-actions commented on Oct 5, 2026

   

   github-actionsBot

   on Oct 5, 2026 – with GitHub Actions

   Contributor

   More actions

   Thanks for updating the issue with a clear description and reproduction steps!
4. opencodejr commented on Oct 27, 2026

   

   opencodejrBot

   on Oct 27, 2026

   Contributor

   More actions

   Thanks for the detailed report. I can reproduce this on 2.0.23, and it still happens on the latest dev (fd8268ea5).

   With the default experimental.subagent_depth (1), I had a simulated model in the primary session start a general subagent:

   • The child session's model request still included the subagent tool, with the full "Available subagents" list in its description.
   • When the child called it, the call failed with Subagent depth limit reached (1). Increase "experimental.subagent_depth" to allow nested subagents.

   The depth limit is only checked when the tool runs. The step that adds the subagent list to the tool description doesn't look at session depth, so the tool is still offered.

   The recording below shows the run, and the script below reproduces it. Until this is fixed, denying subagent in the worker agents' permissions hides the tool, as you found.

   recording.mp4

   Reproduction script (drive.ts)

   // Repro for anomalyco/opencode#53356: the `subagent` tool is still advertised to a
   // child session at the default experimental.subagent_depth (1), and every call fails.
   import { appendFileSync } from "node:fs"
   import { Effect, Stream } from "effect"
   const log = (m: string) => { console.log(m); appendFileSync("/tmp/opencode/repro.log", m + "\n") }
   import { Llm, OpenCodeDriver } from "opencode-drive"
   
   const toolNames = (body: any): string[] =>
     (body?.tools ?? []).map((t: any) => t?.function?.name ?? t?.name).filter(Boolean)
   const text = (body: unknown) => JSON.stringify(body)
   
   export default OpenCodeDriver.use(
     {
       ...(process.env.OC_DEV ? { opencode: { dev: process.env.OC_DEV } } : {}),
       tui: { recording: true },
     },
     ({ ui, llm }) =>
       Effect.gen(function* () {
         yield* llm.serve((request) => {
           const body = request.body as any
           const s = text(body)
           const isChild = s.includes("CHILD_TASK_MARKER")  && !s.includes("PRIMARY_MARKER")
           const tools = toolNames(body)
           log(`[request] ${isChild ? "CHILD" : "PRIMARY"} tools=${JSON.stringify(tools)}`)
           if (isChild) {
             const desc = (body.tools ?? []).find((t: any) => (t?.function?.name ?? t?.name) === "subagent")
             if (desc) log(`[child subagent description]\n${desc?.function?.description ?? desc?.description}`)
             if (s.includes("Subagent depth limit reached")) {
               log("[child] got depth-limit error from subagent tool")
               return Stream.make(Llm.text("Nested delegation failed: depth limit reached."))
             }
             if (tools.includes("subagent"))
               return Stream.make(
                 Llm.toolCall({
                   index: 0,
                   id: "call_nested",
                   name: "subagent",
                   input: { agent: "general", description: "Nested worker", prompt: "Do part of the work" },
                 }),
                 Llm.finish("tool-calls"),
               )
             return Stream.make(Llm.text("Child done without subagent tool."))
           }
           if (s.includes("Nested delegation failed") || s.includes("Child done"))
             return Stream.make(Llm.text("Primary: child finished."))
           return Stream.make(
             Llm.toolCall({
               index: 0,
               id: "call_child",
               name: "subagent",
               input: { agent: "general", description: "Worker task", prompt: "CHILD_TASK_MARKER: split this work up and delegate parts" },
             }),
             Llm.finish("tool-calls"),
           )
         })
         yield* ui.submit("PRIMARY_MARKER delegate this to a worker")
         yield* ui.waitFor("Primary: child finished.", { timeout: 60000 } as any)
         yield* ui.screenshot("final")
       }),
   )

   Run it with opencode-drive: opencode-drive run ./drive.ts. It drives opencode from your PATH; set OC_DEV to the path of an OpenCode checkout to run that from source instead.
5. opencodejr

   added

   reproduced

   on Oct 27, 2026