← Files ProofreadARCHIVED FILE
skills/proofread/references/workflow.md
3.79 KB · Sep 30, 2026 · 23:18 UTC
# Proofreading workflow ## Choose the editing depth - **Grammar-only or light:** correct objective grammar, spelling, punctuation, and obvious word-choice errors. Do not rewrite for style. - **Standard:** apply the grammar pass, then improve clarity, concision, flow, and distracting repetition while preserving the author's voice. - **Direct apply:** the same review standards apply, but explicit user authorization allows writing the reviewed draft over the source without a separate approval round. Never create a translation unless separately requested. Reviewing an existing translation is allowed when requested, but proofreading does not authorize fact-checking, source replacement, citation invention, or substantive rewriting. For source-versus-translation review or optional prose-cleanup companions, read [interoperability.md](interoperability.md). Translation accuracy requires the source; without it, limit the review to target-language fluency and say so. ## Prepare safely For a file, create a temporary directory and copy the source into it with the same extension. Do all work on that copy. For pasted text, a temporary `.md` or `.txt` file is optional but useful when running Harper. Identify the language and spelling dialect from the text or user request. Ask only when choosing incorrectly would materially change the edit. Harper supports English; use it as a baseline rather than as the editor. ## Baseline lint Run: ```bash node scripts/proofread.mjs lint "COPY" --dialect us ``` Supported dialect values are `us`, `uk`, `au`, and `ca`. Exit `0` means Harper found no lint. Exit `1` normally means it found lint findings. Any other exit, or a missing executable, means the baseline did not run. Record the initial findings so the final lint can distinguish pre-existing warnings from regressions. Treat tool output as advice. Product names, domain terms, intentional dialect, and quoted language can be false positives; preserve them and mention material skips in the change summary. ## Human grammar pass Read the entire text. Check agreement, tense, articles, pronoun reference, homophones, punctuation, fragments, run-ons, missing words, and ambiguous modifiers. Prefer the smallest correction that resolves the error. ## Conservative style pass Skip this pass for grammar-only requests. Otherwise: - remove filler, redundant transitions, inflated claims, and formulaic phrasing; - make vague sentences concrete only with details already present in the source; - vary repetitive sentence structure when it distracts; - prefer direct wording when it fits the author's existing tone; - preserve purposeful repetition, technical terminology, accessibility choices, uncertainty, and the writer's punctuation habits; - never add personal experience, opinions, facts, names, numbers, quotations, or sources. Passive voice, adverbs, lists, dashes, and long sentences are not errors by themselves. Change them only when they create a real clarity or style problem in this text. ## Final checks 1. Re-run the same Harper command. Resolve new findings or identify them as false positives. 2. Run `scan` and inspect suspicious invisible Unicode. Do not remove legitimate joiners, directionality controls, or non-breaking spaces blindly. 3. Run `guard ORIGINAL DRAFT`. Any mismatch requires inspection. 4. Review the word diff line by line for meaning drift, dropped qualifications, and accidental formatting changes. 5. Return the corrected pasted text, or show the file diff and wait for approval unless direct application was requested. ## Report Keep the report short. Group meaningful changes under lint, grammar, and style; list material false positives or unresolved ambiguities; state whether the original file was changed. Do not burden the user with every punctuation edit unless they requested a detailed audit.
SHA-256: 7840716d7ec9d6a9aa25ad63743e7096df1884842170576a40864b3f46549cbf