llm-slop-detector 0.8.1 → 0.10.0

package/CHANGELOG.md

@@ -1,5 +1,25 @@
 # Changelog
 
+## [0.10.0](https://github.com/mandakan/llm-slop-detector/compare/llm-slop-detector-v0.9.0...llm-slop-detector-v0.10.0) (2026-05-31)
+
+
+### Features
+
+* add openai pack, citation-artifact tokens, and missing invisibles ([#92](https://github.com/mandakan/llm-slop-detector/issues/92)) ([8b41050](https://github.com/mandakan/llm-slop-detector/commit/8b41050c5c613635a1debf397e0e49ea46497cda))
+* add puffery pack for promotional LLM phrasing ([#94](https://github.com/mandakan/llm-slop-detector/issues/94)) ([f54a711](https://github.com/mandakan/llm-slop-detector/commit/f54a711f6f8b325a04d39adea9adb58f0d5a1201))
+
+## [0.9.0](https://github.com/mandakan/llm-slop-detector/compare/llm-slop-detector-v0.8.1...llm-slop-detector-v0.9.0) (2026-05-30)
+
+
+### Features
+
+* skip NBSP detection in rich-text editors by default ([#79](https://github.com/mandakan/llm-slop-detector/issues/79)) ([ba90929](https://github.com/mandakan/llm-slop-detector/commit/ba90929299e3cc84845901c7091b104ff98e185c))
+
+
+### Bug Fixes
+
+* **deps:** bump vulnerable transitive deps flagged by dependabot ([#95](https://github.com/mandakan/llm-slop-detector/issues/95)) ([90228a7](https://github.com/mandakan/llm-slop-detector/commit/90228a7a468840e06f47ea64be839af90695eb25))
+
 ## [0.8.1](https://github.com/mandakan/llm-slop-detector/compare/llm-slop-detector-v0.8.0...llm-slop-detector-v0.8.1) (2026-04-24)
 
 

package/README.md

@@ -1,6 +1,8 @@
 # LLM Slop Detector
 
 [![Install on VS Code Marketplace](https://img.shields.io/badge/VS%20Code-Marketplace-blue?logo=visualstudiocode)](https://marketplace.visualstudio.com/items?itemName=thias-se.llm-slop-detector)
+[![Install on Chrome Web Store](https://img.shields.io/badge/Chrome-Web%20Store-4285F4?logo=googlechrome&logoColor=white)](https://chromewebstore.google.com/detail/llm-slop-detector/ancpfnfeflffphhpaejhpggfhjmafeid)
+[![Install on Firefox Add-ons](https://img.shields.io/badge/Firefox-Add--ons-FF7139?logo=firefox&logoColor=white)](https://addons.mozilla.org/en-US/firefox/addon/llm-slop-detector/)
 [![License: MIT](https://img.shields.io/github/license/mandakan/llm-slop-detector)](LICENSE)
 
 A VS Code extension and CLI that flag invisible Unicode, AI-style punctuation, and telltale LLM phrases in `markdown` and `plaintext` files. The CLI shares its rule engine with the editor, so local and CI findings stay in sync.
@@ -15,7 +17,12 @@ No install needed -- paste text into the [web playground](https://mandakan.githu
 
 A WebExtension that scans `<textarea>`, text inputs, and `contenteditable` elements (Gmail compose, vanilla rich-text widgets) as you type on any webpage. Findings are highlighted inline with colour-coded marks -- textarea marks are tinted background, contenteditable marks are wavy underlines so they don't disturb reading of styled content. A small "N slop" badge next to each editor opens a per-finding popover with severity, reason, source pack, per-finding "Fix this" buttons, and a "Fix all chars" button that applies every deterministic character replacement (em dash, curly quotes, zero-width, etc.) at once. For contenteditables the fix goes through `execCommand('insertText')` so the host editor's undo stack still works (`Cmd+Z`). Runs entirely in the browser -- no network calls, no telemetry.
 
-**Install from source (until stores are wired up):**
+**Install from a store:**
+
+- **Chrome / Arc / Edge / Brave:** [Chrome Web Store listing](https://chromewebstore.google.com/detail/llm-slop-detector/ancpfnfeflffphhpaejhpggfhjmafeid). Chromium forks accept the same package.
+- **Firefox:** [Firefox Add-ons listing](https://addons.mozilla.org/en-US/firefox/addon/llm-slop-detector/).
+
+**Install from source (for hacking on the rules):**
 
 ```sh
 npm ci
@@ -45,7 +52,7 @@ Turn this off in the options page if you only want to lint your own writing. The
 
 - Flags zero-width, BOM, non-breaking spaces, and other invisible Unicode that hides in text and wrecks diffs
 - Flags AI-style punctuation: em and en dashes, curly quotes, horizontal ellipsis, angle quotes
-- Configurable phrase rules: ~40 built-in core rules plus eleven opt-in packs (`academic`, `cliches`, `fiction`, `claudeisms`, `structural`, `security`, `gemini`, `deepseek`, `llama`, `qwen`, `grok`) totalling 500+ curated regex patterns
+- Configurable phrase rules: ~40 built-in core rules plus thirteen opt-in packs (`academic`, `cliches`, `puffery`, `fiction`, `claudeisms`, `structural`, `security`, `openai`, `gemini`, `deepseek`, `llama`, `qwen`, `grok`) totalling 500+ curated regex patterns
 - Markdown-aware: skips fenced and inline code, link URLs, and YAML frontmatter so technical prose doesn't drown in false positives
 - Inline-ignore comments (`<!-- slop-disable -->`, `<!-- slop-disable-next-line -->`, `<!-- slop-disable-line -->`) for one-off exceptions
 - Hover over any flagged range for the rule selector plus a ready-to-copy `slop-disable-next-line` snippet
@@ -181,13 +188,15 @@ The core list is deliberately conservative: ~40 phrase rules covering the buzzwo
 
 - **`academic`** -- words over-represented in LLM-authored academic writing (`bolster`, `elucidate`, `facilitate`, `showcase`, `noteworthy`, ~90 entries). Severity `hint` so the Problems panel stays usable. Derived from [`berenslab/llm-excess-vocab`](https://github.com/berenslab/llm-excess-vocab) (MIT).
 - **`cliches`** -- general LLM cliche vocabulary (`captivating`, `pinnacle`, `galvanize`, journey/landscape/symphony metaphors). Derived from [`nanxstats/llm-cliches`](https://github.com/nanxstats/llm-cliches) (MIT).
+- **`puffery`** -- promotional / editorializing puffery: `rich cultural heritage`, `nestled in the heart of`, `boasts`, `enduring legacy`, `leaves an indelible mark`, `deeply rooted`, `diverse array of`, copula-avoidance flourishes. Original content.
 - **`fiction`** -- fiction and creative-writing tells (breath hitched, heart hammering, shivers down spine, chestnut eyes, LLM-cliche character names). **Includes adult-fiction markers.** Derived from [`SicariusSicariiStuff/SLOP_Detector`](https://github.com/SicariusSicariiStuff/SLOP_Detector) (Apache-2.0).
 - **`claudeisms`** -- Claude-specific mannerisms: sycophantic openers, consent-theater phrasing, "important to note that" hedges. Derived from SLOP_Detector.
 - **`structural`** -- structural LLM tells: "not X but Y" negation pivots, sycophantic line openers, "in this section we'll" meta-commentary, "at the end of the day" closers. Original content.
-- **`security`** -- LLM-weaponized invisibles above the BMP: tag characters (U+E0020-U+E007F, used in ASCII-smuggler prompt injection) and variation selectors (U+FE00-U+FE0E + U+E0100-U+E01EF, used for arbitrary-data smuggling in emoji and CJK). Severity `error`. Skips U+FE0F (emoji presentation selector) to avoid false positives. Opt in if you copy-paste LLM output into files you don't fully trust.
+- **`security`** -- LLM-weaponized invisibles above the BMP: tag characters (U+E0020-U+E007F, used in ASCII-smuggler prompt injection) and variation selectors (U+FE00-U+FE0E + U+E0100-U+E01EF, used for arbitrary-data smuggling in emoji and CJK), plus invisible math operators (U+2061-U+2064), deprecated format controls (U+206A-U+206F), the combining grapheme joiner (U+034F), Mongolian free variation selectors (U+180B-U+180D), and the Braille blank (U+2800). Severity `error` (`warning` for the few code points with rare legitimate use). Skips U+FE0F (emoji presentation selector) to avoid false positives. Opt in if you copy-paste LLM output into files you don't fully trust.
 
 Model-family packs flag mannerisms specific to one vendor's models. Mix and match:
 
+- **`openai`** -- OpenAI GPT / ChatGPT: citation-markup artifact leaks (`oaicite`, `oai_citation`, `contentReference`, `turn0search0`-style web-tool tokens), AI-identity and knowledge-cutoff disclaimers (`as an AI language model`, `knowledge cutoff`), refusal boilerplate (`I cannot fulfill that request`), and stock opener/closer/hedge scaffolding (`Certainly!`, `here's a comprehensive overview`, `I hope this helps`). Original content.
 - **`gemini`** -- Google Gemini / Bard: breakdown-happy structure (`here's a breakdown`, `let's break it down`), analogy openers (`think of it as`, `imagine a`), `I hope this helps` closers, and identity / knowledge-cutoff / professional-advice disclaimers.
 - **`deepseek`** -- DeepSeek V3 / R1: `<think>` tag leaks, R1 reasoning tics leaking into final output (`Wait,`, `Hmm,`, `Let me reconsider`, `the user is asking`), DeepSeek special-token leaks (`<|begin_of_thought|>`, `<|EOT|>`).
 - **`llama`** -- Meta Llama: `as an AI` identity disclaimers, Llama-Guard refusal boilerplate (`I cannot provide`, `it would not be appropriate`), and Llama2 / Llama3 chat-template token leaks (`[INST]`, `<|eot_id|>`, `<<SYS>>`, `<|python_tag|>`).

package/builtin-packs/grok.json

@@ -1,6 +1,6 @@
 {
   "name": "pack:grok",
-  "version": "0.1.0",
+  "version": "0.2.0",
   "description": "xAI Grok mannerisms: conversational 'look,' openers, forced edginess, Hitchhiker's Guide references, and Elon/X name-drops.",
   "phrases": [
     { "pattern": "^\\s*look,?\\s", "reason": "Grok opener" },
@@ -35,6 +35,7 @@
     { "pattern": "\\bdon'?t panic\\b", "reason": "Grok Hitchhiker's reference" },
     { "pattern": "\\bhitchhiker'?s guide\\b", "reason": "Grok Hitchhiker's reference" },
     { "pattern": "\\bmostly harmless\\b", "reason": "Grok Hitchhiker's reference" },
+    { "pattern": "\\bgrok_card\\b", "reason": "Grok citation-markup artifact leak", "severity": "warning" },
     { "pattern": "\\bi'?m grok\\b", "reason": "Grok identity disclaimer" },
     { "pattern": "\\bi am grok\\b", "reason": "Grok identity disclaimer" },
     { "pattern": "\\bas grok,?\\b", "reason": "Grok identity disclaimer" },

package/builtin-packs/openai.json

@@ -0,0 +1,34 @@
+{
+  "name": "pack:openai",
+  "version": "0.1.0",
+  "description": "OpenAI GPT / ChatGPT tells: citation-markup artifact leaks (oaicite, contentReference), AI-identity and knowledge-cutoff disclaimers, refusal boilerplate, and stock opener/closer/hedge scaffolding. Original content -- regexes hand-authored to match commonly-observed stock phrasing; no third-party rule set incorporated.",
+  "phrases": [
+    { "pattern": "\\boaicite\\b", "reason": "ChatGPT citation-markup artifact leak" },
+    { "pattern": "\\boai_citation\\b", "reason": "ChatGPT citation-markup artifact leak" },
+    { "pattern": "\\bcontentReference\\b", "reason": "ChatGPT citation-markup artifact leak", "severity": "warning" },
+    { "pattern": ":contentReference\\[oaicite", "reason": "ChatGPT citation-markup artifact leak", "severity": "warning" },
+    { "pattern": "\\bturn\\d+(search|news|view|image|forecast|finance)\\d+\\b", "reason": "ChatGPT web-tool citation token leak", "severity": "warning" },
+    { "pattern": "\\bas an? (AI|artificial intelligence)(,| developed| created| trained| language model| assistant)", "reason": "GPT identity disclaimer", "severity": "warning" },
+    { "pattern": "\\bas a large language model\\b", "reason": "GPT identity disclaimer", "severity": "warning" },
+    { "pattern": "\\bas of my last (knowledge |training )?(update|cutoff|knowledge update)\\b", "reason": "GPT knowledge-cutoff disclaimer", "severity": "warning" },
+    { "pattern": "\\bknowledge cut[- ]?off\\b", "reason": "GPT knowledge-cutoff disclaimer", "severity": "warning" },
+    { "pattern": "\\bmy (training data|training|knowledge)( only)? (includes|extends|goes|is limited|cuts off|stops)\\b", "reason": "GPT knowledge-cutoff disclaimer", "severity": "warning" },
+    { "pattern": "\\bI (don'?t|do not) have (access to |the ability to )?(real[- ]time|live|up[- ]to[- ]date|current|personal)\\b", "reason": "GPT capability disclaimer", "severity": "warning" },
+    { "pattern": "\\bI (can'?t|cannot|am unable to|am not able to) (fulfill|comply with|assist with|help with) (that|this) request\\b", "reason": "GPT refusal boilerplate", "severity": "warning" },
+    { "pattern": "\\bI'?m (sorry|afraid),? but (as|I)\\b", "reason": "GPT refusal boilerplate" },
+    { "pattern": "^\\s*Certainly[!,]", "reason": "GPT opener" },
+    { "pattern": "^\\s*Sure(?: thing)?[!,]", "reason": "GPT opener" },
+    { "pattern": "^\\s*Sure,? (here'?s|here is|I can|let me)\\b", "reason": "GPT opener" },
+    { "pattern": "\\bGreat question[!.]", "reason": "GPT opener" },
+    { "pattern": "\\bI'?d be (happy|glad) to (help|assist)\\b", "reason": "GPT opener" },
+    { "pattern": "\\bhere'?s a (comprehensive|detailed|quick|brief|step[- ]by[- ]step) (overview|breakdown|rundown|guide|look)\\b", "reason": "GPT framing" },
+    { "pattern": "\\blet me break (this|it) down( for you)?\\b", "reason": "GPT framing" },
+    { "pattern": "\\bI hope (this|that) (helps|information helps|answers your question)\\b", "reason": "GPT closer" },
+    { "pattern": "\\blet me know if (you (have|need)|there'?s|you'?d like|you want)\\b", "reason": "GPT closer" },
+    { "pattern": "\\b(feel free|don'?t hesitate) to (ask|reach out|let me know)\\b", "reason": "GPT closer" },
+    { "pattern": "\\bis there anything else (I can|you'?d like)\\b", "reason": "GPT closer" },
+    { "pattern": "\\bgenerally speaking,?\\b", "reason": "GPT hedge" },
+    { "pattern": "\\bit'?s (important|worth|essential) to (note|remember|keep in mind|mention) that\\b", "reason": "GPT hedge" },
+    { "pattern": "\\bdepending on (the|your) (context|specific|particular|needs|situation)\\b", "reason": "GPT hedge" }
+  ]
+}

package/builtin-packs/puffery.json

@@ -0,0 +1,19 @@
+{
+  "name": "pack:puffery",
+  "version": "0.1.0",
+  "description": "Promotional / editorializing puffery common in LLM prose: 'rich cultural heritage', 'nestled in the heart of', 'boasts', 'enduring legacy', 'leaves an indelible mark', and copula-avoidance flourishes. Original content -- regexes hand-authored, no third-party rule set incorporated.",
+  "phrases": [
+    { "pattern": "\\brich (cultural|literary|artistic|musical|architectural|natural|historical|spiritual) heritage\\b", "reason": "puffery" },
+    { "pattern": "\\b(stands|serves|stood|served) as a testament\\b", "reason": "puffery" },
+    { "pattern": "\\bleaves? an indelible mark\\b", "reason": "puffery" },
+    { "pattern": "\\bvaluable insights?\\b", "reason": "LLM filler" },
+    { "pattern": "\\bnestled (in|between|among|amid|along|within|atop|beside|deep)\\b", "reason": "travel-writing puffery" },
+    { "pattern": "\\benduring (legacy|appeal|charm|popularity|influence|presence|significance)\\b", "reason": "puffery" },
+    { "pattern": "\\bvibrant (community|culture|tapestry|atmosphere|scene|hub|ecosystem|nightlife|array|mix|blend)\\b", "reason": "puffery" },
+    { "pattern": "\\bboasts? (a|an|the|its|some|several|numerous|over|more than|impressive|stunning|an array|a range)\\b", "reason": "copula-avoidance puffery" },
+    { "pattern": "\\bdeeply rooted\\b", "reason": "puffery" },
+    { "pattern": "\\bsetting the stage for\\b", "reason": "LLM transition cliche" },
+    { "pattern": "\\b(a |an )?diverse array of\\b", "reason": "puffery" },
+    { "pattern": "\\bbreathtaking (views?|scenery|beauty|landscapes?|vistas?)\\b", "reason": "puffery" }
+  ]
+}

package/builtin-packs/security.json

@@ -1,7 +1,7 @@
 {
   "name": "pack:security",
-  "version": "0.1.0",
-  "description": "LLM-weaponized invisibles above the BMP: tag characters (ASCII-smuggler prompt injection) and variation selectors (arbitrary-data smuggling). Severity: error.",
+  "version": "0.2.0",
+  "description": "LLM-weaponized invisibles above the BMP: tag characters (ASCII-smuggler prompt injection), variation selectors (arbitrary-data smuggling), invisible math operators, and deprecated/zero-width format controls used to hide instructions or smuggle data. Severity: error (warning for chars with rare legitimate use).",
   "chars": [
     {"char":"󠀁","name":"LANGUAGE TAG (U+E0001)","severity":"error","replacement":""},
     {"char":"󠀠","name":"TAG \" \" (U+E0020)","severity":"error","replacement":""},
@@ -354,6 +354,21 @@
     {"char":"󠇬","name":"VARIATION SELECTOR-253 (U+E01EC)","severity":"error","replacement":""},
     {"char":"󠇭","name":"VARIATION SELECTOR-254 (U+E01ED)","severity":"error","replacement":""},
     {"char":"󠇮","name":"VARIATION SELECTOR-255 (U+E01EE)","severity":"error","replacement":""},
-    {"char":"󠇯","name":"VARIATION SELECTOR-256 (U+E01EF)","severity":"error","replacement":""}
+    {"char":"󠇯","name":"VARIATION SELECTOR-256 (U+E01EF)","severity":"error","replacement":""},
+    {"char":"⁡","name":"FUNCTION APPLICATION (U+2061)","severity":"error","replacement":""},
+    {"char":"⁢","name":"INVISIBLE TIMES (U+2062)","severity":"error","replacement":""},
+    {"char":"⁣","name":"INVISIBLE SEPARATOR (U+2063)","severity":"error","replacement":""},
+    {"char":"⁤","name":"INVISIBLE PLUS (U+2064)","severity":"error","replacement":""},
+    {"char":"͏","name":"COMBINING GRAPHEME JOINER (U+034F)","severity":"warning","replacement":""},
+    {"char":"","name":"INHIBIT SYMMETRIC SWAPPING (U+206A)","severity":"error","replacement":""},
+    {"char":"","name":"ACTIVATE SYMMETRIC SWAPPING (U+206B)","severity":"error","replacement":""},
+    {"char":"","name":"INHIBIT ARABIC FORM SHAPING (U+206C)","severity":"error","replacement":""},
+    {"char":"","name":"ACTIVATE ARABIC FORM SHAPING (U+206D)","severity":"error","replacement":""},
+    {"char":"","name":"NATIONAL DIGIT SHAPES (U+206E)","severity":"error","replacement":""},
+    {"char":"","name":"NOMINAL DIGIT SHAPES (U+206F)","severity":"error","replacement":""},
+    {"char":"᠋","name":"MONGOLIAN FREE VARIATION SELECTOR ONE (U+180B)","severity":"warning","replacement":""},
+    {"char":"᠌","name":"MONGOLIAN FREE VARIATION SELECTOR TWO (U+180C)","severity":"warning","replacement":""},
+    {"char":"᠍","name":"MONGOLIAN FREE VARIATION SELECTOR THREE (U+180D)","severity":"warning","replacement":""},
+    {"char":"⠀","name":"BRAILLE PATTERN BLANK (U+2800)","severity":"warning","replacement":""}
   ]
-}
+}

package/extension-browser/manifest.json

@@ -1,7 +1,7 @@
 {
   "manifest_version": 3,
   "name": "LLM Slop Detector",
-  "version": "0.8.1",
+  "version": "0.10.0",
   "description": "Flags invisible Unicode, AI-style punctuation, and telltale LLM phrases as you type -- or in any page you're reading. Local only.",
   "permissions": [
     "storage"

package/out/core/rules.js

@@ -4,7 +4,7 @@ exports.BUILTIN_PACKS = exports.LOCAL_RULES_FILENAME = void 0;
 exports.parseSeverityOverrides = parseSeverityOverrides;
 exports.loadRules = loadRules;
 exports.LOCAL_RULES_FILENAME = '.llmsloprc.json';
-exports.BUILTIN_PACKS = ['academic', 'cliches', 'fiction', 'claudeisms', 'structural', 'security', 'gemini', 'deepseek', 'llama', 'qwen', 'grok'];
+exports.BUILTIN_PACKS = ['academic', 'cliches', 'puffery', 'fiction', 'claudeisms', 'structural', 'security', 'openai', 'gemini', 'deepseek', 'llama', 'qwen', 'grok'];
 function parseSeverity(s, fallback) {
     switch (s) {
         case 'error': return 'error';

package/package.json

@@ -2,7 +2,7 @@
   "name": "llm-slop-detector",
   "displayName": "LLM Slop Detector",
   "description": "Highlights invisible Unicode, AI-style punctuation, and telltale LLM phrases in markdown and plain text.",
-  "version": "0.8.1",
+  "version": "0.10.0",
   "publisher": "thias-se",
   "engines": {
     "vscode": "^1.95.0"
@@ -75,10 +75,12 @@
             "enum": [
               "academic",
               "cliches",
+              "puffery",
               "fiction",
               "claudeisms",
               "structural",
               "security",
+              "openai",
               "gemini",
               "deepseek",
               "llama",
@@ -88,10 +90,12 @@
             "enumDescriptions": [
               "Words over-represented in LLM-authored academic writing (derived from berenslab/llm-excess-vocab, MIT)",
               "General LLM cliche adjectives/nouns/verbs (derived from nanxstats/llm-cliches, MIT)",
+              "Promotional / editorializing puffery: 'rich cultural heritage', 'nestled in the heart of', 'boasts', 'enduring legacy', 'leaves an indelible mark', copula-avoidance flourishes. Original content.",
               "Fiction and creative-writing LLM tells, including NSFW markers (derived from SicariusSicariiStuff/SLOP_Detector, Apache-2.0)",
               "Claude-specific mannerisms and sycophantic consent-theater phrasing (derived from SicariusSicariiStuff/SLOP_Detector, Apache-2.0)",
               "Structural LLM tells: 'not X but Y', sycophantic openers, meta-commentary",
-              "Flags LLM-weaponized invisibles above the BMP: tag chars (ASCII-smuggler prompt injection) and variation selectors (arbitrary-data smuggling). Severity: error.",
+              "Flags LLM-weaponized invisibles above the BMP: tag chars (ASCII-smuggler prompt injection) and variation selectors (arbitrary-data smuggling), plus invisible math operators and deprecated format controls. Severity: error/warning.",
+              "OpenAI GPT / ChatGPT tells: citation-markup artifact leaks (oaicite, contentReference), AI-identity and knowledge-cutoff disclaimers, refusal boilerplate, and stock opener/closer/hedge scaffolding",
               "Google Gemini / Bard tells: breakdown-happy structure, 'I hope this helps' closers, identity and knowledge-cutoff disclaimers",
               "DeepSeek V3/R1 tells, notably <think> tags and R1 reasoning tics ('Wait,', 'Hmm,', 'Let me reconsider') leaking into output",
               "Meta Llama tells: 'as an AI' disclaimers, Llama-Guard refusal boilerplate, and Llama2/Llama3 chat-template token leaks ([INST], <|eot_id|>, etc.)",