Terminal Lag Recovery

Terminal Lag Recovery

Recovery Workflow

Use this skill when interactive typing in the Codex terminal has become slow or stalled and the user wants local recovery.

1. Inspect for zeroclaw processes before killing anything:

   ps -axo pid,ppid,stat,etime,pcpu,pmem,command | rg -i 'zeroclaw|codex'
   

2. Run the bundled recovery script:

   python3 ~/.codex/skills/terminal-lag-recovery/scripts/recover_zeroclaw.py
   

3. If the sandbox cannot read the process table or send signals, rerun the same command with escalated approval. Use a concise justification such as: "Do you want to allow targeted zeroclaw process cleanup to recover the lagging terminal UI?"

4. Report what happened:

   • Say how many zeroclaw processes were found and terminated.
   • If none were found, say the cleanup was a no-op and mention any obvious high-CPU process only if you inspected it.
   • Do not terminate codex cli, Terminal, shell, or unrelated system agents as part of this skill.

Escalation

Start with SIGTERM. Use SIGKILL only when the user explicitly asks for a stronger cleanup or when a previously matched zeroclaw process survives SIGTERM and the user has approved escalation:

python3 ~/.codex/skills/terminal-lag-recovery/scripts/recover_zeroclaw.py --kill

Notes

• Treat "zeroclaw threads" as OS processes unless a managed-agent API explicitly exposes closable zeroclaw sessions.
• A pkill -TERM -if zeroclaw fallback is acceptable when the script is unavailable, but the script is preferred because it prints the exact matched PIDs.
• If process inspection shows no zeroclaw process, do not invent a cleanup. State that no matching processes were present.