cyberneticlibrary

Systematically decompose and execute work

ulw-loopskillsetup L261,462
code-yeongyu/oh-my-openagent
What it does

Execute goal-like work with systematic decomposition and manual QA

Best for

Long-running durable work with checkpoints when context resets and manual QA are acceptable overhead.

Inputs
  • · user goal or task
  • · evidence of completion (tmux/HTTP/browser/computer-use)
Outputs
  • · .omo/ulw-loop ledger with checkpointed work
  • · evidence-bound success criteria
Requires
  • · Codex spawn_agent, wait_agent, close_agent
  • · optional subagent workers
Preconditions

.omo/ulw-loop/ state writable; clear success criteria defined

Failure modes
  • · State lost between context resets (use .omo/ulw-loop/ledger.jsonl recovery)
  • · QA skipped despite success output
  • · Subagent child created without TASK:/DELIVERABLE/SCOPE headers
Trust signals
  • · Evidence recorded through CLI only after cleanup receipts available
  • · Ledger-based recovery survives context loss
  • · Delegation discipline (fork_turns: none, wait_agent, polling) prevents deadlock