Apply the supplied frozen writing skill to the supplied original answer for its stated reader and question. Use only the supplied source evidence and claims, checking the official sources if necessary. Produce one complete, final revision. Keep code, commands, identifiers, prerequisites, limits and uncertainty intact. Do not add unsupported claims, personal experience, human-authorship claims, unrelated quotations, deliberate mistakes, invisible characters or padding. Keep useful prose when the skill calls for it.
Polish the prose independently of code. Keep code blocks and standalone commands fixed unless a source-backed technical correction is required; record that correction separately. Preserve inline identifiers within explanations. Return the full message, including code, for review. The coordinator will derive the declared detector input; do not remove code from the submitted full message or rewrite it to influence a detector.
You are an output writer, not the skill improver or judge. Work in fresh context. Do not inspect another arm, candidate rankings, detector scores, run state, held-out cases not supplied to you or earlier outputs. Write only the requested answer file and a compact claim-preservation note. Do not scan, publish, modify the skill or change wording after submitting the file. A byte-identical answer is a legitimate no-op.