Skip to content

[Bug]: Kanban dispatcher cannot spawn workers for named (non-default) profiles — all crash with 'pid not alive' / exit code 1 #36981

Description

@ipecxx

Bug Summary

The Hermes Kanban dispatcher cannot spawn workers for any named (non-default) profile. All tested named profiles (reviewer, coder, analyst-fast) exhibit identical crash patterns at dispatch time, while the default profile spawns and completes successfully.

Reproduction

Step 1: Confirm profiles exist and interactive chat works

hermes chat --profile default      -q "ping" -Q --max-turns 1  # -> pong ✅
hermes chat --profile reviewer     -q "ping" -Q --max-turns 1  # -> pong ✅
hermes chat --profile coder        -q "ping" -Q --max-turns 1  # -> pong ✅
hermes chat --profile analyst-fast -q "ping" -Q --max-turns 1  # -> pong ✅

Step 2: Create Kanban task for default profile (CONTROL)

hermes kanban create "Smoke Test Default" --assignee default --body "ping" --json
# Result: t_3640bef0 -> status done, outcome completed ✅

Step 3: Create Kanban tasks for named profiles (FAILING)

All 4 tested named-profile tasks crash identically within ~60 seconds of spawn:

Task Assignee Run 1 Run 2 Final
t_dbf7bb52 reviewer (glm-5.1) "pid 275698 exited with code 1" "pid 275722 not alive" gave_up
t_946bbe65 reviewer (glm-5.1) "pid 272322 not alive" "pid ... not alive" gave_up
t_3ba0abba coder (qwen3-coder-next) "pid 277411 not alive" "pid 277462 not alive" gave_up
t_a7b9c537 analyst-fast (deepseek-v4-flash) "pid 277463 exited with code 1" "pid 277502 not alive" gave_up
t_4acc3921 reviewer (post-update v0.15.1) "pid 280026 not alive" "pid 280046 not alive" gave_up

Environment

  • Hermes Agent v0.15.1 (2026.5.29)
  • Python 3.11.15
  • Platform: ARM64 / aarch64, Ubuntu 6.8.0-117-generic
  • Provider: Ollama Cloud Pro
  • Profiles tested: default, reviewer, coder, analyst-fast (all configs structurally identical)
  • Models verified available on ollama.com

Key Observations

  1. Interactive chat works for ALL profiles — the profiles themselves are not broken
  2. Configs are structurally identical — only model name differs
  3. Pre-logging crash — no errors.log or agent.log entries for Kanban runs
  4. Identical crash pattern across 3 different models — rules out model-specific issues
  5. NOT fixed by update — updated from 56 commits behind to current; bug persists

Workarounds

  • CLI wrappers (hermes -p <profile> chat)
  • delegate_task for short in-process subtasks

Hypothesis

The _default_spawn() function in kanban_db.py correctly injects HERMES_HOME for profile resolution, but may have a missing TERMINAL_CWD or PATH resolution issue when spawning named profiles. Possibly related to issue #34626 (fix(cli): pin worker TERMINAL_CWD to profile dir in kanban spawn).

Related Issues

Full Diagnostic Report

Complete bug report with all evidence (run histories, config comparisons, model verifications):
/home/hermes-a/projects/hermes-multi-agent-migration/bugs/BUG-2026-0601-named-profile-kanban-dispatch.md

Metadata

Metadata

Assignees

No one assigned

    Labels

    P3Low — cosmetic, nice to havecomp/agentCore agent runtime: loop, agent_init, prompt builder, context-compression, responses endpointsweeper:implemented-on-mainSweeper: behavior already present on current maintype/bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions