Ver Fonte

Extend pattern 11 to repeated sentence openings (v2.9.2) (#209)

Cover repeated sentence openings while preserving deliberate anaphora. Keep the numbered pattern count at 33.
Som Tripathi há 1 mês atrás
pai
commit
eda497e31a
3 ficheiros alterados com 14 adições e 6 exclusões
  1. 1 1
      .claude-plugin/plugin.json
  2. 2 1
      README.md
  3. 11 4
      SKILL.md

+ 1 - 1
.claude-plugin/plugin.json

@@ -2,7 +2,7 @@
   "$schema": "https://json.schemastore.org/claude-code-plugin-manifest.json",
   "name": "humanizer",
   "description": "Remove signs of AI-generated writing from text, making it sound more natural and human. Based on Wikipedia's \"Signs of AI writing\" guide.",
-  "version": "2.9.1",
+  "version": "2.9.2",
   "author": {
     "name": "blader",
     "url": "https://github.com/blader"

+ 2 - 1
README.md

@@ -131,7 +131,7 @@ Rewrites follow a no-fabrication rule: they never add facts, names, dates, or ci
 | 8 | **Copula avoidance** | "serves as... features... boasts" | "is... has" |
 | 9 | **Negative parallelisms / tailing negations** | "It's not just X, it's Y", "..., no guessing" | State the point directly |
 | 10 | **Rule of three** | "innovation, inspiration, and insights" | Use natural number of items |
-| 11 | **Synonym cycling** | "protagonist... main character... central figure... hero" | "protagonist" (repeat when clearest) |
+| 11 | **Synonym cycling / repeated openings** | "protagonist... main character... hero" or "She noted... She noted... She filed..." | Settle on one referent; merge the repeated run |
 | 12 | **False ranges** | "from the Big Bang to dark matter" | List topics directly |
 | 13 | **Passive voice / subjectless fragments** | "No configuration file needed" | Name the actor when it helps clarity |
 
@@ -207,6 +207,7 @@ Rewrites follow a no-fabrication rule: they never add facts, names, dates, or ci
 
 ## Version History
 
+- **2.9.2** - Extended pattern #11 to cover the opposite of synonym cycling: consecutive sentences that all open on the same subject, usually a pronoun, with the merge, subject-shift, and action-first fixes and a narrative example. Added a deliberate-anaphora false-positive guard so rhetorical repetition is left alone (fixes #206). No change to the 33 patterns.
 - **2.9.1** - Improved distribution and portability: removed nonportable frontmatter and tool preapprovals, made global installation the documented default, added package validation, and removed the duplicated long-form example from the runtime prompt. No change to the 33 patterns.
 - **2.9.0** - Added a no-fabrication rule: rewrites may not invent facts, names, dates, or citations not present in the source, and every example that modeled invented specifics was re-cut to use only source information (fixes #187). Replaced paragraph-count parity with an information-over-shape rule, made a user's voice sample outrank the em dash ban, and added invocation modes (pasted text / file / embedded). No change to the 33 patterns.
 - **2.8.3** - Moved the skill version from the unsupported top-level frontmatter key to `metadata.version` for Agent Skills and Claude compatibility. No change to the 33 patterns.

+ 11 - 4
SKILL.md

@@ -9,7 +9,7 @@ description: |
   voice, negative parallelisms, and filler phrases.
 license: MIT
 metadata:
-  version: "2.9.1"
+  version: "2.9.2"
 ---
 
 # Humanizer: Remove AI Writing Patterns
@@ -145,12 +145,18 @@ When voice is appropriate, avoid uniform sentence structures, bloodless neutrali
 **After:**
 > The event includes talks and panels. There's also time for informal networking between sessions.
 
-### 11. Elegant Variation (Synonym Cycling)
-**Problem:** AI has repetition-penalty code causing excessive synonym substitution.
-**Before:**
+### 11. Elegant Variation and Repeated Sentence Openings
+**Problem:** AI has repetition-penalty code causing excessive synonym substitution. The same machinery misses in the other direction in narrative prose, where consecutive sentences all open on the same subject, usually a pronoun, and nothing varies where the sentence starts. Both are one defect: the model is managing repetition by rule instead of by ear. Cure over-variation by settling on a single referent. Cure under-variation by merging the sentences, by giving the subject role to something other than the character, or by opening on the action so the pronoun arrives later.
+**Before (synonym cycling):**
 > The protagonist faces many challenges. The main character must overcome obstacles. The central figure eventually triumphs. The hero returns home.
 **After:**
 > The protagonist faces many challenges but eventually triumphs and returns home.
+**Before (repeated openings):**
+> She noted the door. She noted the lock on it. She filed both away.
+**After:**
+> She noted the door and its lock, then filed both away.
+
+The fix is not banning the repeated word. A run of three sentences becoming one is what removes the tell; the survivor may still start with "She."
 
 ### 12. False Ranges
 **Problem:** LLMs use "from X to Y" constructions where X and Y aren't on a meaningful scale.
@@ -367,6 +373,7 @@ A clean human writer can hit several of the patterns above without any AI involv
 - **Curly quotes alone.** macOS, Word, Google Docs, and most CMSes auto-curl by default. Curly quotes only count when stacked with other tells.
 - **Em dashes alone.** Many editors and journalists use them often. Em dashes are evidence only when paired with formulaic sales-y rhythm.
 - **One short emphatic sentence.** Humans use clipped sentences to land a point. Flag staccato drama only when several short fragments appear in a row and inflate the tone.
+- **Deliberate anaphora.** Repeating a sentence opening on purpose is an old device, and good prose uses it to build cadence or pressure ("She came. She saw. She conquered."). Flag a repeated opening only when the run does no rhetorical work and reads as the model failing to vary rather than a writer choosing.
 - **"Honestly" or "look" mid-sentence.** These are ordinary in casual writing. The tell is the standalone theatrical opener, not the word itself.
 - **Unsourced claims.** Most of the web is unsourced. Lack of citations doesn't prove anything.
 - **Correct, complex formatting.** Visual editors and templates produce clean output without any AI.