goal-verifiedlisted

# goal-verified A loop-engineering wrapper for `/goal`. It exists to fix one failure mode: an agent that **builds and verifies its own work always says "looks good"**, because it cannot see its own blind spots — so you inherit the bugs it didn't point out. The fix is to put an independent "no"-sayer between the build and the commit, and to pin a falsifiable bar *before* building so the verifier has something concrete to check. regin does not own the agent loop, so this is a procedure you (the agent) must follow, not something enforced from outside. Follow it in order. Do not skip the STOP. ## Why this works (read once) - `/goal` alone stops when the model *runs out of ideas* (`prompt_input_exit`), not when a goal is *verified met*. That is why bugs slip through. - "Make the UI good" is unfalsifiable — there is nothing to check, so "done" collapses to a vibe. The roadmap converts it into checkable items anchored to standards the repo **already holds** (the same skills/engines the hooks enforce), so you are never inventing a bar. - Division of labour: the **program gathers** candidates (deterministic, high recall — never misses a relevant standard) and **you refine** them (precision — only judgment grounded in the real code can tell *relevant* from merely *related*). Skipping the refine (step 1.5) hands the build a noisy bar; that is the difference between a roadmap that helps and one that distracts. ## Two ways to run this - **Inline mode** (default, si