improvelisted

# ASK Improve Structured, evidence-first codebase audit that produces executable implementation plans. Also covers focused refactoring and simplification passes. Runs once to audit and plan; re-invoke `execute`, `reconcile`, or `next` modes for follow-through. ## When to use Use when: - you want a structured audit (correctness, security, performance, tech debt, deps, DX, docs, direction, user-facing text language/locale consistency) - you have audit output and need executable implementation plans - you need to execute, reconcile, or publish those plans as issues - the user asks for a refactor, cleanup, simplification, or deduplication Skip when: - the problem is a single known bug (use `debugging`) - you only need a quick code review (use `code-review`) ## Audit flow 1. **Recon**: detect repo layout, conventions, commands, existing docs/ADRs, intent signals (TODOs, flags, stubs, PRDs, roadmap) 2. **Audit**: run category audits, including code-smell scanning — each finding is evidence-grounded (`file:line`) 3. **Plan**: convert findings into self-contained, executor-ready plans with drift checks, verification gates, hard boundaries, and escape hatches 4. **Follow-through**: `execute <plan>`, `reconcile`, `next`, `branch` For significant or release-sensitive work, act as the independent audit role: inspect callers and the complete diff, challenge ownership, ambiguity, fallbacks, determinism, compatibility, and test invariants. Return concrete evidence, not generic qualit