Next Task Recommendation
Identify the highest-priority next task based on current project state.
1. Gather State (Minimal Reads)
Read only:
- •
gitignore/worklog.doing.md— full (short by design) - •
gitignore/worklog.todo.md— full (short by design) - •Project plan if it exists (
gitignore/plan*.md) — priority/phase section only, not the full doc
Then run a single directory listing to see what's implemented:
bash
ls -1 src/*/
Do not read worklog.done.md — doing + todo + directory state is sufficient.
2. Determine Current State
From what you read, answer:
- •Is there anything in progress (doing file)? → If yes, recommend finishing it first.
- •What does the plan say is the current priority? → Use the plan's own phase/priority structure, not hardcoded assumptions.
- •What is not yet implemented in
src/? → Cross-reference plan priorities against actual files.
If no plan document exists, infer priorities from todo order and directory gaps.
3. Priority Logic
- •In-progress tasks exist → recommend completing the current task
- •Todo has items → recommend the top item that matches the current phase/priority
- •Todo is empty → compare plan priorities against
src/and suggest what's missing - •Phase boundary → if all current-phase work looks done, ask: "Start next phase?"
- •Missing tests → if an implementation file exists with no corresponding test, flag it
4. Output Format
code
## Next Task ### Current State - Phase: [derived from plan or inferred] - In progress: [task or "none"] - Phase completion: [rough estimate, e.g. "3 of ~7 tasks"] ### Recommended Task **[filename or task name]** #### What to do [What needs to be implemented — derived from plan doc, not hardcoded] #### Reference docs [Link to relevant plan section if found] #### Getting started \```bash [command or initial snippet if applicable] \``` #### When done \``` worklog done [task] complete next \```
Notes
- •Derive phase structure and file expectations from the plan document — never assume a fixed structure
- •If the plan has changed, reflect that; do not rely on stale mental models
- •Keep reads to: doing + todo + one plan section + one
ls