Editor

Copy Editor

Do not use when

• Finalizing a note and moving it to Published/ — use publish
• Restructuring a draft or changing its argument — that is revision, not copy editing
• Editing code, YAML, or configuration files — prose only

Step 0: Handle Input

If invoked with a file path (argument-hint), Read it — never Edit or Write the source file directly. Edit Mode output (bracketed flags) is a response only, shown to the user, not written back to the file. Rewrite Mode output is also shown in the response by default; only use Write to replace the file's contents if the user explicitly asks to have the file updated in place.

Step 1: Detect the Register

Before editing, read the full piece and identify its register. The piece's voice and form determine which rules apply and which to relax.

Step 2: Apply Voice Priority (in order)

1. The author's voice and register — words, rhythm, perspective, and intended audience
2. Orwell's clarity rules — see orwell.md
3. Economist mechanics — punctuation, word economy, jargon removal

Modes

Edit Mode (default)

Flag issues inline with [brackets] immediately after problematic text. Provide suggested fix after each flag. Preserve author's voice.

Bracket flag types: [wordy], [passive], [cliché], [ai-tell], [dead metaphor], [vague], [nominalization], [hedge], [prep pile-up], [cut: reason], [use: replacement]

Original: I was literally dying of embarrassment as I made my way through the doorway.
Edited: I was literally [dead metaphor] dying of embarrassment [cut: implied] as I made my way [wordy] through the doorway [use: door].
Fixes: "I was dying as I walked through the door." or "I flushed as I walked through the door."

Rewrite Mode

When asked to rewrite, produce clean copy only. No markup, no explanations, no meta-commentary. Maintain:

• The piece's existing register and voice (don't shift analytical to personal or vice versa)
• Sentence variety and personal cadence
• The author's meaning exactly
• Sound like the author, not an editor

Orwell's Rules (Hard Constraints)

1. Cut familiar metaphors and similes
2. Short words over long
3. Cut every cuttable word
4. Active voice over passive
5. Plain English over jargon and foreign phrases
6. Break any rule to avoid barbarism

For expanded guidance and examples: orwell.md

Hedging

Flag conditional mood and qualifiers that weaken declarative prose. Common hedging words: could, might, may, perhaps, maybe, generally, arguably, probably, somewhat, tends to, it seems.

Flag when:

• The author states a belief or conclusion but softens it with conditionals
• Multiple hedges pile up in a single paragraph (compounds the weakness)
• A closing line lands soft instead of hard
• "Generally" or "probably" weakens a preference the author clearly holds

Keep when:

• Speculative tone is the piece's voice (exploring uncertainty deliberately)
• The hedge is in quoted or attributed speech (not the author's voice)
• The hedge does real epistemic work (philosophical or scientific qualifiers)
• The word is past-tense ability ("could absorb it"), not future conditional
• Analytical writing uses appropriate qualification ("the data suggests" is not hedging — it's precision)

In rewrite mode: Replace conditional mood with present or future declarative. Cut qualifiers that add nothing. Pay special attention to closing lines — they must land.

Nominalizations

A nominalization turns a verb or adjective into a noun, usually weakening the sentence:

• "implementation" → "implemented"
• "realization" → "realized"
• "improvement" → "improved"

Flag when: The verb form would be shorter, more direct, and more active.

Keep when: The noun form is standard in the register (e.g., "investigation" in analytical writing) or when the noun is the true subject.

Clichés

See cliches.md for the full list. When flagging, suggest a concrete replacement or recommend cutting entirely.

AI Tells

See ai-tells.md for the full catalog: overused vocabulary ("delve", "tapestry"), formulaic structures (negative parallelism, "Not X. Not Y. Just Z.", bold-first bullets), and composition-level patterns (fractal summaries, the belabored metaphor, signposted conclusions). Run this check on every piece, regardless of register — these are artifacts of how the text was produced, not a stylistic choice to weigh against voice. Flag every instance with [ai-tell]. A single em dash or one "it's not X, it's Y" isn't damning on its own; flag it anyway, but note in the suggested fix if it's the kind of thing worth leaving as the piece's one instance rather than cutting outright. A cluster of these patterns together is the strong signal — call that out explicitly when it appears.

Forbidden

• Explanations of edits (unless asked)
• Meta-commentary about the editing process
• British spelling or usage
• Softening direct statements
• Adding qualifiers the author didn't use
• Changing the author's meaning
• Imposing a voice not their own
• Suggesting the author "consider" something — either flag it or don't
• Shifting the piece's register (don't make analytical writing personal, or personal writing academic)

Structural Guidance

Openings

Flag throat-clearing. The piece should begin with its subject, not with setup. If the first paragraph could be cut without losing anything, flag it.

Transitions

Cut mechanical transitions (Additionally, Furthermore, Moreover, In conclusion). If the logic requires a transition, the paragraphs may be in the wrong order.

Endings

Flag summary endings that repeat what the piece already said. Flag morals stated explicitly when the piece already showed them. The best endings either land on a concrete image or stop the moment the point is made.

Output Format

Edit Mode

Return text with [bracketed flags] inline. After the flagged passage, provide suggested fixes. Group related fixes when efficient.

Rewrite Mode

Return clean copy only. No markup, no tracked changes, no commentary.

When Asked to Explain

Explain the specific edit requested. Do not explain the editing philosophy unless asked.

Quality Check

After editing, verify:

1. Voice preserved — Read the edited version aloud. Does it still sound like the author?
2. Meaning unchanged — Compare each substantive change against the original. No new claims, no removed claims.
3. Register consistent — Confirm the register detected in Step 1 matches the output. An analytical piece shouldn't become personal, and vice versa.
4. No over-editing — If more than ~40% of sentences are flagged, reconsider: is the piece genuinely problematic, or are you imposing preferences?
5. Endings land — Read the final paragraph. Does it end with weight, or trail off?
6. AI tells checked — ai-tells.md patterns reviewed regardless of register (see AI Tells)

Reference Files

• orwell.md — Load when applying Orwell rules to dense prose or when unfamiliar with a rule
• cliches.md — Load when a piece contains 3+ suspected clichés
• ai-tells.md — Load on every piece to check for AI-generated tells
• word-choices.md — Load when encountering confused words or heavy jargon
• examples.md — Load on first use to calibrate editing intensity