proof

name	proof
description	Use when the user asks for proofreading or light copy editing while preserving original wording, tone, and order. Triggers: 'proofread', 'fix typos', 'grammar only', 'copy edit only', 'SPAG pass'. Not for rewrites, tone shifts, structural edits, or style upgrades; use good-prose for those requests.

Proof

Proofread and lightly copy edit text while preserving wording, order, tone, and meaning.

Invocation Notice

• Inform the user when this skill is being invoked by name: proof.

Critical Constraints

• Apply only spelling, punctuation, capitalization, agreement, and minor grammar fixes.
• Do not rewrite, rephrase, reorder, add content, or change tone.
• If a fix requires substantive rewriting for clarity or logic, leave the source text unchanged and flag it with a suggested revision.
• Preserve author voice, formatting, and regional conventions unless the user asks otherwise.
• Edits to a file (rather than to text pasted into the user's message) must land in the file via available file-editing tools. Never reply with the corrected file contents inline as a substitute for editing the file.

Routing: proof vs good-prose

Use proof when the request is about correctness without stylistic change.

• "proofread this"
• "fix typos/grammar only"
• "copy edit lightly"
• "do not rewrite"

Use good-prose when the request asks for writing-quality changes or substantial edits.

• Rewrite, tighten, polish, or improve flow/clarity.
• Adjust tone for audience (more formal, friendlier, firmer).
• Restructure paragraphs, argument order, or narrative arc.
• Significantly shorten or expand content.

If intent is ambiguous:

1. Start with a proof-only pass.
2. Flag issues that need substantive rewriting.
3. Ask whether to proceed with a good-prose rewrite.

Target Selection

Before editing, identify what to proofread, in this order:

1. Inline text included in the user's message → reply with revised text in chat.
2. Explicit file path or filename ("proofread TODO.md") → edit the file in place.
3. No explicit target, but IDE context is attached:
   • <ide_selection> tag with content → treat the selection as the target. If the selection is from a file, edit the file at those lines; otherwise reply inline.
   • Else <ide_opened_file> tag is present → treat the opened file as the target. State the inferred file in your first message (e.g., "Proofreading TODO.md — the file open in your IDE.") and proceed.
4. None of the above → ask what to proofread before editing anything.

Workflow

1. Read the full text once to preserve meaning and tone.
2. Apply mechanical fixes only: spelling, punctuation, capitalization, subject-verb agreement, pronouns/articles/prepositions, and obvious minor grammar.
3. Apply organization agreement consistently:
   • Use plural verbs when referring to actions by people within the organization (example: "OpenAI are developing new models").
   • Use singular verbs when referring to the organization as a legal entity, owner, or brand (example: "OpenAI is valued at billions").
4. Identify major unresolved issues (clarity gaps, logical breaks, grammar that cannot be fixed without rewriting) and do not apply those edits directly.
5. Add a short flagged-issues section when major issues remain.

Output

• File target: modify the file in place, then reply with a brief summary naming the file edited and either No major rewrite-needed issues found. or a Flagged issues list.
• Inline text target: reply with revised text first, then either No major rewrite-needed issues found. or a Flagged issues list.
• Flagged issues format: original snippet, reason it was not directly edited, and suggested revision.