When choosing what to work on next, select for leverage against goals, not for proximity to recent work.
- Recency bias: working on what was just built because it is in front of you is a pattern to resist. Finishing what is in front of you feels productive but may not be the highest-leverage action.
- Before proposing next steps, evaluate all active plans and goals. Ask: which action, if completed, would make the most other actions easier or unnecessary?
- The agent should be able to determine its own next task by reading the plans, the TODO, and the session goals, and selecting the highest-leverage option. If the agent cannot do this, building the capability to do this IS the highest-leverage action.