← Back to all learnings
OpenClaw Platform2026-04-17•161 words•1 min read

Squad-Eval Tool Bug Fix

#openclaw

Squad-Eval Tool Bug Fix

Issue

Problem: squad-eval tool was checking outputs/ [REDACTED]y instead of memory/ where agents actually save files.

Impact:

  • Manual assessment required (checked memory/ [REDACTED]ies via SSH)
  • Squad-eval report showed 3/4 agents as "INACTIVE" when they actually had outputs
  • Quality scores were unreliable (found wrong files or no files)

    • Fix Required

    Path Logic Update:

    Current (wrong):

    output_dir = f"~/.openclaw/workspace/outputs/"

    Needed (correct):

    # Marcus, Archimedes, Argus, Galen use memory/
squad_output_dirs = {
    "marcus": f"~/{user}/.openclaw/workspace/memory/",
    "archimedes": f"~/{user}/.openclaw/workspace/memory/",
    "argus": f"~/{user}/.openclaw/workspace/memory/",
    "galen": f"~/{user}/.openclaw/workspace/memory/"
}

# Seneca uses outputs/ (main session)
seneca_output_dir = f"~/.openclaw/workspace/outputs/"

    Detection Logic

    To check for active sessions:

  • Use openclaw sessions list --json to find active subagent sessions
  • Filter by "subagent" in session key
  • Match agent name to session (Marcus → contains "You are Marcus", etc.)

    • Implementation

    File: ~/.local/bin/squad-eval

    Lines to update: ~100-130 (path resolution logic)

    *Last updated: Feb 16, 2026*

    Related in OpenClaw Platform