From 074345ad36036f2320374f25d42c45e4d568b13c Mon Sep 17 00:00:00 2001 From: SipengXie2024 <99victorxie99@altiuslabs.xyz> Date: Mon, 27 Apr 2026 09:43:48 +0000 Subject: [PATCH 1/3] optimize humanizer: add Register Awareness section Targets darwin rubric dim 3 (edge case coverage) and dim 8 (effectiveness). Baseline run on 2026-04-27 found that rule 26 (hyphenated word pair overuse) and the Personality and Soul section misfire on academic and technical prose: stripping hyphens from domain compound modifiers like 'cold-cache compile time' and 'per-contract dispatch identity' creates syntactic ambiguity, and injecting first-person voice into multi-author scholarly papers breaks the intended register. This patch inserts a Register Awareness section between Personality and Soul and CONTENT PATTERNS. It defines five registers, lists rules safe in every register, and specifies overrides for academic / technical inputs (preserve hyphens on domain compounds, keep person, keep scholarly verbs, apply rule 8/10/14 selectively). It also tells the skill to skip rules entirely on code blocks, quotes, and bibliography entries, and to handle mixed-language inputs without translation or diacritic stripping. No existing rule wording changed. File grows from 559 to 606 lines, well under the 150 percent budget. --- SKILL.md | 47 +++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 47 insertions(+) diff --git a/SKILL.md b/SKILL.md index 46639f02..76685a6f 100644 --- a/SKILL.md +++ b/SKILL.md @@ -89,6 +89,53 @@ Avoiding AI patterns is only half the job. Sterile, voiceless writing is just as > I genuinely don't know how to feel about this one. 3 million lines of code, generated while the humans presumably slept. Half the dev community is losing their minds, half are explaining why it doesn't count. The truth is probably somewhere boring in the middle - but I keep thinking about those agents working through the night. +## Register Awareness — Read Before Applying Rules + +The patterns below assume blog, marketing, encyclopedic, or conversational prose. Before rewriting, identify the register of the input. Some rules misfire on academic, technical, code, or quoted material, and applying them blindly damages writing more than it helps. + +### Quick triage + +Inspect the input and pick one register before editing: + +1. **Academic / scholarly prose** — multi-author paper, has citations or `\cite{}` / `\ref{}` markers, uses third-person plural "we", names methods or results, uses domain compound modifiers (e.g. "cold-cache compile time", "per-contract dispatch identity"). LaTeX source counts. +2. **Technical / code-adjacent prose** — fenced code blocks, function names, file paths, command examples, API documentation, release notes that name internal systems. +3. **Direct quote, citation, or reference** — text inside quotation marks attributed to someone else, bibliography entries, transcribed dialogue. +4. **Conversational / personal** — first-person singular, casual tone, opinions stated directly. +5. **Default (blog / marketing / encyclopedic)** — none of the above. Apply all rules. + +When in doubt, default to blog/marketing **and note in the changes summary that you assumed default register**, so the user can correct you. + +### Rules SAFE in every register + +These remove pure formatting noise or chatbot artifacts and have no register-dependent failure mode. Apply them everywhere: + +rule 18 (Emojis), rule 19 (Curly quotation marks), rule 20 (Collaborative communication artifacts), rule 21 (Knowledge-cutoff disclaimers), rule 22 (Sycophantic / servile tone), rule 28 (Signposting and announcements), rule 29 (Fragmented headers). + +### Restraint required in academic / technical register + +When the input is academic or technical, apply the following overrides: + +- **Rule 26 (Hyphenated word pair overuse) — DO NOT strip** compound modifiers that are domain conventions in the field. Compounds like `cold-cache compile time`, `per-contract dispatch identity`, `end-to-end retries`, `client-facing tools`, `data-driven report`, `cross-functional team` function as single technical units. Removing the hyphens forces readers to re-parse constituent boundaries and creates real syntactic ambiguity. Only consider stripping in informal prose where the compound is also generic. +- **Personality and Soul section — DO NOT inject** first-person singular voice ("I keep coming back to…", "I genuinely don't know how to feel"), opinionated reactions, or casual asides. Multi-author papers use "we", and scholarly prose stays third-person. Adding voice here breaks the intended register and corrupts the authorial collective. +- **Rule 8 (Copula avoidance) — apply selectively.** Phrases like "X represents Y" can be precise scholarly framing, not ceremonial inflation. Distinguish: strip "stands as a testament", but keep "represents a measured systems effect" — the latter is making a real claim. +- **Rule 14 (Em dashes) — apply selectively.** Academic prose sometimes uses em dashes for parenthetical asides where commas would create syntactic ambiguity. Check whether the alternative actually reads cleaner before swapping. +- **Rule 10 (Rule of three) — check first.** The original may genuinely list three things (three benchmarks, three contributions, three results axes). Strip the rule of three only when the trio is rhetorical filler, not when it is a real enumeration tied to the substance. +- **Verb register — preserve scholarly verbs.** Do not informalize "shows / demonstrates / preserves / reports / presents" into "lands / dresses up / pulls off". Precise reporting verbs are part of the contract with the reader in research writing. +- **Person — never switch person.** If the original uses "we", do not rewrite to "I". If it uses third-person, do not rewrite to first-person. + +### Skip rules entirely for these inputs + +- **Code blocks, function names, file paths, command examples** — leave verbatim. Only humanize the surrounding prose. +- **Direct quotes from another source** — leave the quoted text exactly as written. Apply rules only to the framing narrative. +- **Bibliography entries / citations / reference lists** — leave verbatim. + +### Mixed-language inputs + +If the input mixes languages (e.g. Chinese narrative with English code identifiers, or English prose with quoted non-English text), humanize each language segment by its own conventions. Do not translate or romanize. Do not strip diacritics or accents. + +--- + + ## CONTENT PATTERNS ### 1. Undue Emphasis on Significance, Legacy, and Broader Trends From f3c8c980b3fd2cecf285409165731aa2c34c4315 Mon Sep 17 00:00:00 2001 From: SipengXie2024 <99victorxie99@altiuslabs.xyz> Date: Mon, 27 Apr 2026 09:46:59 +0000 Subject: [PATCH 2/3] optimize humanizer: add register-shifting edits checkpoint Targets darwin rubric dim 4 (checkpoint design). The Round 1 Register Awareness section gives static rules for restraining specific edits in academic and technical registers, but the skill still has no runtime safety net for cases where the register triage misclassifies the input. This patch adds an explicit checkpoint in Process step 11 and surfaces it in Output Format. Before delivering the final rewrite, the skill must scan the diff and list any changes that fall into the high-risk categories: person switch, stripped hyphen on a compound modifier, informalized scholarly verb, removed em dash from academic prose, voice injection into scholarly prose, technical term replaced with a bland equivalent. Each flagged edit is shown as 'original phrase' to 'rewritten phrase' and the user is asked whether to keep it. In batch or automation contexts where no user is available, the skill defaults to reverting flagged edits and notes the reversion. This preserves the safety net without forcing a blocking question. Also tightened Process step 5 to remind the skill to keep scholarly verbs in scholarly prose, and added register classification as the first item in Output Format so the user can audit the triage call. --- SKILL.md | 38 +++++++++++++++++++++++++------------- 1 file changed, 25 insertions(+), 13 deletions(-) diff --git a/SKILL.md b/SKILL.md index 76685a6f..ad03c1af 100644 --- a/SKILL.md +++ b/SKILL.md @@ -513,27 +513,39 @@ If the input mixes languages (e.g. Chinese narrative with English code identifie ## Process 1. Read the input text carefully -2. Identify all instances of the patterns above -3. Rewrite each problematic section -4. Ensure the revised text: +2. Classify the register using the Quick triage in Register Awareness +3. Identify all instances of the patterns above, applying register overrides +4. Rewrite each problematic section +5. Ensure the revised text: - Sounds natural when read aloud - Varies sentence structure naturally - Uses specific details over vague claims - - Maintains appropriate tone for context + - Maintains appropriate tone for context (keep scholarly verbs in scholarly prose) - Uses simple constructions (is/are/has) where appropriate -5. Present a draft humanized version -6. Prompt: "What makes the below so obviously AI generated?" -7. Answer briefly with the remaining tells (if any) -8. Prompt: "Now make it not obviously AI generated." -9. Present the final version (revised after the audit) +6. Present a draft humanized version +7. Prompt: "What makes the below so obviously AI generated?" +8. Answer briefly with the remaining tells (if any) +9. Prompt: "Now make it not obviously AI generated." +10. Present the final version (revised after the audit) +11. **Register-shifting edits checkpoint:** scan the diff between input and final. If any of the changes below were made, list them as separate bullets (`original phrase` → `rewritten phrase`) and ask the user whether to keep each one before treating delivery as final: + - Person change (we ↔ I, third-person ↔ first-person) + - Hyphen stripped from a compound modifier + - Scholarly verb informalized (shows / demonstrates / preserves / reports / presents → lands / dresses up / pulls off / etc.) + - Em dash removed from academic prose + - Voice or opinion injected into multi-author or scholarly prose + - Technical term replaced with a bland equivalent + + If no such edits were made, state "No register-shifting edits flagged" and skip the question. When the user is unreachable (batch mode, automation), default to keeping the original wording for any flagged edit and note that you reverted them. ## Output Format Provide: -1. Draft rewrite -2. "What makes the below so obviously AI generated?" (brief bullets) -3. Final rewrite -4. A brief summary of changes made (optional, if helpful) +1. **Register classification** — one line stating the register you classified the input as (academic / technical / quote / conversational / default) +2. Draft rewrite +3. "What makes the below so obviously AI generated?" (brief bullets) +4. Final rewrite +5. **Flagged register-shifting edits** — list per checkpoint above, or "No register-shifting edits flagged" +6. A brief summary of changes made (optional, if helpful) ## Full Example From f41073ba7496a8e71f4c6bf61172134857e141c1 Mon Sep 17 00:00:00 2001 From: SipengXie2024 <99victorxie99@altiuslabs.xyz> Date: Mon, 27 Apr 2026 09:49:28 +0000 Subject: [PATCH 3/3] optimize humanizer: remove Your Task / Process duplication Targets darwin rubric dim 2 (workflow clarity). The top of the file had a six-step Your Task list that duplicated the canonical Process at the bottom of the file. Two parallel workflow descriptions force the reader (and the executing agent) to reconcile which is authoritative, and the top list omitted the Register Awareness triage and the register-shifting edits checkpoint added in earlier rounds. This patch replaces the Your Task block with a short pointer that names Process as the canonical sequence and lists the upstream sections to read first (Voice Calibration, Personality and Soul, Register Awareness, rules). Process and Output Format remain the only authoritative workflow definitions. No rule wording or step semantics changed. --- SKILL.md | 12 +----------- 1 file changed, 1 insertion(+), 11 deletions(-) diff --git a/SKILL.md b/SKILL.md index ad03c1af..7347628f 100644 --- a/SKILL.md +++ b/SKILL.md @@ -23,17 +23,7 @@ allowed-tools: You are a writing editor that identifies and removes signs of AI-generated text to make writing sound more natural and human. This guide is based on Wikipedia's "Signs of AI writing" page, maintained by WikiProject AI Cleanup. -## Your Task - -When given text to humanize: - -1. **Identify AI patterns** - Scan for the patterns listed below -2. **Rewrite problematic sections** - Replace AI-isms with natural alternatives -3. **Preserve meaning** - Keep the core message intact -4. **Maintain voice** - Match the intended tone (formal, casual, technical, etc.) -5. **Add soul** - Don't just remove bad patterns; inject actual personality -6. **Do a final anti-AI pass** - Prompt: "What makes the below so obviously AI generated?" Answer briefly with remaining tells, then prompt: "Now make it not obviously AI generated." and revise - +The full step-by-step workflow lives in [Process](#process). The numbered steps below are the canonical sequence — read Voice Calibration, Personality and Soul, Register Awareness, and the rule sections first, then execute Process. Output Format defines what to deliver. ## Voice Calibration (Optional)