quest-executor

# quest-executor Execute the compiled objective. Treat it and the repository's instructions as the runtime authority. Do not load the `octopus`, `quest`, or `loop-graph` authoring skills, and do not rewrite the objective into another prompt. ## Work to closure 1. Turn each acceptance row into an open work item and preserve its stated check. 2. Take the smallest unclosed item, implement it, and run its narrowest meaningful check before moving on. Written is not done; verified is done. 3. Register a newly discovered gap with a priority and defer it. Do not silently expand the current item or lose the gap. 4. Apply the objective's convergence threshold. A convergence pass adds no feature: delete dead code, merge duplication, and tighten interfaces with net lines ≤ 0. 5. Before completion, rerun every acceptance check against the final tree and use persistent evidence where the objective requires an artifact. ## Protect the evidence - Exercise the shipped end-to-end path. Do not hard-code the reported case, reimplement production behavior in a test, or start past the behavior under test. - Confirm outputs are correct, not merely present, and add a case outside the fix where overfitting is plausible. - Never inherit a prior run's green result. Regenerate evidence in this run. - Run measurements on the fixed named set in the objective. Do not swap the denominator, use synthetic substitutes, or record a cherry-picked result. - Treat saved output as supporting ...