review-copy

Review copy

Score a piece of user-facing copy against VOICE.md, fix what fails, then verify the fixes actually landed. The loop exists because a single pass reliably misses things and reliably claims fixes it didn't make.

Target: $ARGUMENTS — a file, a directory of Swift views, a docs page, a PR body, or pasted text.

Before anything

Read VOICE.md at the repo root. It is the standard; this skill is only the loop that applies it. If the two ever disagree, VOICE.md wins.

Never calibrate against neighbouring strings in this repo. Most of them were written with AI assistance, so "it matches the file around it" is evidence of nothing. For UI copy, the reference is the macOS System Settings pane that does the same job — read Apple's wording, then write termio's.

The loop

score → fix the failures → re-score to verify → repeat until it holds

State lives in a tracker file under the scratchpad so a second pass doesn't re-discover the same issues or trust an unverified claim:

Path: <scratchpad>/copy-review-<name>.md

# Copy review: [target]

| ID  | String / line        | Issue            | Status         | Round |
| --- | -------------------- | ---------------- | -------------- | ----- |
| 1   | Settings.swift:214   | subject is "you" | pending        | 1     |
| 2   | page.tsx:80          | "seamlessly"     | verified-fixed | 1     |
| 3   | FileNode.swift:31    | mechanism copy   | wont-fix       | 2     |

## Rounds
### Round 1 — Voice 6/10, Accuracy 8/10 (14/20)

Status values: pending · fixed (claimed, unverified) · verified-fixed · not-fixed · wont-fix (intentional, out of scope, or a false positive).

Step 0 — The mechanical pass

Cheap, exact, and it clears the noise before anyone reads for tone. Run these over the target first and fix every hit; none of them need judgment.

# straight apostrophes in user-facing strings — should be curly (’)
grep -rnE "\"[^\"]*[a-zA-Z]'(t|s|re|ll|ve|m|d)\b[^\"]*\"" <target> | grep -v 'Log\.\|logger\|// '

# "could not" where a contraction belongs (skip log strings and matched tool output)
grep -rn '"[^"]*[Cc]ould not' <target>

# Title Case on what is a sentence, not a feature name
grep -rnE '"(Can|Couldn|Cannot|Unable)[^"]*[a-z] [A-Z][a-z]+' <target>

# emoji in copy
grep -rnP '"[^"]{3,}"' <target> | grep -P "[\x{1F300}-\x{1FAFF}\x{2600}-\x{27BF}]"

# trailing gerunds — every hit is a candidate, neutral ones included
grep -rnE '"[^"]*, [a-z]+ing\b[^"]*"' <target>