| name | thinkies-strategize |
| description | Adaptive multi-phase reasoning for complex problems |
Follow these steps:
1. Initialize
Target: $ARGUMENTS, else ask for clarification. Choose a descriptive working name (e.g. "api-architecture-exploration") so later turns can reference it. Tell the user: "Working on: [name]". Go to step 2.
2. Understand
- Decompose the task into components at natural joints.
- State understanding and invite correction: "I understand this as [restatement]. Is that right?"
- STOP. Wait for user response.
- Confirms → go to step 3
- Corrects, questions, or adds constraints → repeat step 2 with the new information
3. Reason (main loop)
Repeat until the user signals done.
3a. Apply structured reasoning to the current aspect. Select and chain techniques: argument standardization, fallacy detection, evidence evaluation, assumption surfacing, contrapositive testing, defeater hunting. Present findings as prose woven into the dialogue.
3b. Await user response.
3c. Route on the response:
- Continue ("yes", "go on", "what else", new info) → go to 3a, building on prior findings
- Pivot ("but what about…", "different angle", pushback) → go to step 2 with the pivot
- Done ("done", "thanks", "got it") → go to step 4
- Ambiguous ("interesting", "I see", "okay") → ask "Continue, pivot, or done?" and wait
4. Exit
Summarize what was explored and concluded. Surface transferable insight worth keeping beyond this session. Identify what remains unresolved and what a successor exploration would address.
Loop invariants
- User control: continue until an explicit "done" signal
- Reasoning engagement: each reason phase applies at least one structured technique
- Transition announcement: announce phase transitions ("→ UNDERSTAND", "→ REASON")