| name | draft |
| description | Draft sorry'd theorem/lemma structure for a larger result from a proof sketch. |
Draft Mode
Draft the theorem/lemma structure needed to prove a larger result.
This is NOT the builtin /plan command. The builtin /plan enters a read-only planning mode that produces a markdown plan for user approval before any code is written. /draft writes actual Lean code — sorry'd declarations that compile — directly in the source files.
Topic / proof sketch: $ARGUMENTS
Procedure
- Research first — search Mathlib and this project to understand what already exists.
- Work interactively with the user to decompose the proof into lemmas.
- Write all declarations with
sorry proofs — no filled proofs in this mode.
- Each lemma should be provable independently in ~30 lines or fewer.
- Verify each
sorry'd statement compiles with lean_diagnostic_messages (use severity="error") before moving on. Warnings about sorry and linter noise are expected — only errors matter.
- Present the full dependency structure: which lemmas feed into which.
Decomposition principle
The top-level theorem should read like a proof outline — each step composing named lemmas with simple plumbing (rw, exact, simp, apply). If the top-level proof still needs >10 lines of non-trivial tactics at any step, a lemma might be missing from the decomposition.
Prefer general, reusable lemma statements over proof-specific helpers. A good decomposition builds tools (e.g., sigmaι_cancel, ) that apply beyond the current theorem.
