sloplint 0.5.0 → 0.7.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 74566267ccb2ee1d4caec464d3098809e24f232f75872a7df2ac61ef7d583153
-  data.tar.gz: 4e3b6196aeb6604d212d00be14032e530a644d4250820424774244fe7298fa9e
+  metadata.gz: 33eff57c67530d0745b1ba5009a97128929a73ec27ae09afaba8d9057f0f997b
+  data.tar.gz: 3b76104840a3c2ab471651cbdf18dd4a9027707aaf100bc901430c7f9e121216
 SHA512:
-  metadata.gz: 41e707ec52643d3f57abe04867d399fef77c8e567378394e4c1e0e2ca541d8aefaa2cb908b9d0e419fc9af9f0a1aedb55adce9ef87fc187628be4a90641a4809
-  data.tar.gz: ed1b87c4a32490d582100343d2e7a4b114067737de3761fa3fb768e669f398e69b8c4e3ce4cb33d3138702d533818f5dd53b2b778d8736f848ca71c8166e61f8
+  metadata.gz: 4d1a47eae97395fa37dd1fa50b9a4894ccae70064d15542abdce0197cf715956d23c6ba75bee7bb4f56bc6ada8986485aff5bb1e06c9ecbb1ac6b763705588e1
+  data.tar.gz: cb1ba2ed058855336b71bdc6e10a743e021a5d5d85e0207b8e0564b0fcd9076c9c45eb021eee6fc15a4eac92709b3440ebb6e1487da3c8981853ff37ebd01419

data/CHANGELOG.md

@@ -3,7 +3,335 @@
 All notable changes to this project are documented here. Format loosely
 follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
-## [Unreleased]
+## [0.7.0] - 2026-09-15
+
+### Added
+
+- `--strict` runs the full catalog, including the five rules that ship off
+  by default. Before this, the only way to turn one on was to name it in
+  `--select`, which also turned off every other rule unless you listed
+  those too. `--ignore` still subtracts on top of `--strict`, and
+  `--select` is unchanged.
+- `ellipsis-closer` (structure, info) flags the long-sentence-then-short-closer
+  kicker `mic-drop-closer` catches, but built on the opposite verb shape: the
+  closer's verb phrase is elided down to a bare auxiliary with no object,
+  "The agent did." rather than a quantifier plus a full verb. No subject list
+  is needed — the tell is the missing verb, not the subject — so "That",
+  "This" and "It" are not excluded the way `mic-drop-closer` excludes them.
+  A negative lookahead drops a closer that still holds a wh-word, since that
+  marks a subordinate clause with its own complement rather than an elided
+  one. Reuses `mic-drop-closer`'s long-sentence prefix.
+- `actually-not-x`: the adverb and a trailing "…, not X" in one clause, as in
+  "The disclosure actually covered two incidents, not one". Either marker
+  alone is ordinary technical English and neither is matched on its own; the
+  tell is the two of them doubled up, correcting an alternative the reader was
+  never offered. The narrowing is structural, not a word list: the comma in
+  front of "not" has to be the first comma of its clause and has to follow a
+  word, so a fronted setup ("Despite the name, …") and a parenthetical both
+  drop it. Nothing crosses a line break, so two rows of a table cannot weld
+  into one correction.
+- `real-x-real-y` (rhetorical-tic, info) flags the same "real" used twice
+  attributively in one sentence, in front of two different nouns: "it can
+  trigger real API calls or hold real credentials". Bare "real" is an
+  ordinary word, so the narrowing is the repetition itself, not a noun
+  list. A hyphen on either side takes it out ("real-time", "non-real"), a
+  backreference requires the second noun to differ from the first, and a
+  short closed list drops the fixed senses of this register (real time,
+  real-world, real numbers, real user monitoring, real estate, real money).
+- `phrase-echo` (rhetorical-tic, info, off by default) flags a three-word
+  phrase that comes back within a few hundred words, the model reusing a
+  phrase it has just minted. Each word must be four characters or more and
+  lowercase-led, and one of them six letters with nothing but letters, so "in
+  order to", "couldn't have been" and "New York City" never match. The window
+  does not cross into a list item or a table row. Off by default: a term of
+  art repeats because it must, and in reference prose the pattern fires on
+  names by the thousand per million words.
+- `trailing-restatement` (structure, info, off by default) flags the tail
+  that says the sentence again in other words: ", which means …", ", which
+  is to say …", ", meaning …", and four participles that hang a result off
+  the sentence (", making it easier", ", allowing us to", ", giving them
+  more", ", leaving you with"). The connective is all a pattern can see, and the
+  same connective states a real consequence in careful prose, so the rule
+  runs only when selected.
+
+### Fixed
+
+- `check`'s options now work without the command word: `sloplint --markdown -`
+  and `sloplint --strict -` scan, as the README has said `sloplint -` does.
+  Before, the global parser rejected any option it did not own with
+  `invalid option`, so `check` had to be spelled out whenever an option
+  came first. An option no command knows still exits 2.
+- `--markdown` now blanks `<!-- -->` comment blocks along with fenced code,
+  inline code and URLs. Planning notes kept in comments never render, so
+  every rule that fired inside one was a permanent false positive.
+- `rule-of-three`'s message, rationale, README and SPEC entries said it
+  flagged three parallel comma items. The pattern requires single words in
+  the closing two slots, so a triad of phrases never matched; the text now
+  says that.
+
+## [0.6.0] - 2026-09-08
+
+The catalog grows from 50 rules to 75, and sloplint ships as a Claude Code
+plugin as well as a gem.
+
+### Added
+
+- sloplint now ships as a Claude Code plugin as well as a gem. Add the
+  marketplace `benjaminjackson/sloplint` and install `sloplint`; the plugin
+  carries the Ruby source with it, so nothing is installed from a package
+  index and it works where `gem install` cannot reach the network — Claude
+  Cowork above all. The plugin adds one skill, `/sloplint:check`, which runs
+  the scanner and reports what it returns. The skill refuses to assess prose
+  on its own when the scanner cannot run, because a review with no scan
+  behind it reads exactly like one with a scan behind it.
+- `not-by-x-but-by-y`: the corrective frame on a repeated preposition, "not
+  by luck, but by design". The copula rules cannot see it because nothing
+  anchors "not" but a verb or a dash, so the anchor is the preposition, which
+  must repeat after "but". Ships at `info`: Thoreau and Melville write it too.
+
+### Fixed
+
+- `check` now reads files and stdin as UTF-8 whatever the locale says. A
+  sandbox with no `LANG` set — Claude Cowork's, among others — leaves Ruby's
+  default external encoding at US-ASCII, and the first em dash in a draft then
+  failed the whole scan with `invalid input: invalid byte sequence in
+  US-ASCII`. Prose is the only input sloplint takes, so UTF-8 is the
+  assumption rather than the locale's guess. Genuinely invalid UTF-8 is still
+  exit 2.
+
+### Changed
+
+- `thats-the-whole` now takes `value` and `fix` after "that's the whole",
+  alongside `point`, `game`, `thing`, `deal` and `story`. The contracted
+  closer is the one an agent writes in technical prose ("That's the whole
+  value of the `info` tier", "That's the whole fix"), and `is-the-whole-x`
+  sees only the uncontracted "is", so the sentence slipped both rules. The
+  uncontracted "That is the whole fix." moves with it, from `info` under
+  `is-the-whole-x` to `warning`. The two new nouns must end the sentence or
+  the paragraph, or run into one of the words a closer trails off on, so
+  `thats-the-whole` leaves "value chain", "value-add" and "fix list" alone
+  however the line wraps (`is-the-whole-x` still reads the uncontracted
+  "That is the whole value chain" at `info`, as it did), and a curly
+  apostrophe now counts as a contraction. Both rules interpolate one `WHOLE_CLOSERS`
+  fragment, so the lists cannot drift, and a spec walks every noun through
+  every form the two wrappers treat differently. Neither noun appears after
+  "that's the whole" in 1.55M words of public-domain prose.
+
+- Empty or whitespace-only input
+ is now exit 2 with `empty input: nothing to
+  check in ...`, instead of exit 0. Reporting a scan of nothing as a clean
+  scan is the same trap a mistyped rule id sets, and gets the same answer. The
+  text is tested before `--markdown` blanks code and URLs, so a file holding
+  only a fenced code block still exits 0; and only an all-empty set of sources
+  is an error, so one empty file among several named ones is left alone.
+
+- The executable moved from `bin/sloplint` to `exe/sloplint`. claude.ai
+  rejects a plugin with a top-level `bin/` directory, which would block
+  organization distribution. `gem install sloplint` is unaffected.
+
+- `exe/sloplint` now checks `RUBY_VERSION` before loading anything and aborts
+  with an explanation, rather than raising `NoMethodError` from `Data.define`.
+  The Ruby macOS ships at `/usr/bin/ruby` is 2.6 and hits this.
+
+### New rules
+
+- `from-x-to-y-chain` (rhetorical-tic, warning) flags two or more
+  comma-separated "from X to Y" spans in a row: "from guessing to measuring,
+  from hoping to knowing". Each span names two poles and nothing between
+  them, and stacking them sweeps across a change without describing it. Two
+  human shapes are skipped: the relay, where each span starts where the last
+  ended ("from the egg to the worm, from the worm to the fly"), and the
+  reduplication, where a span has the same word at both ends ("from hummock
+  to hummock"). Operands must open with a letter, so a list of ranges ("from
+  1990 to 1995, from 1997 to 2001") is not a chain. With those out it flags once in 2.4M words of pre-2022
+  Hacker News (a geographic sweep) and once in 1.25M words of public-domain
+  prose (Joyce).
+
+- `one-x-one-y` (rhetorical-tic, warning) flags three or more "one X" items
+  in a comma chain that stands on its own: "One owner, one repository, one
+  weekly prune." The word does no counting; it sets a rhythm. The chain must
+  open a sentence or follow a colon, because after a verb ("the flat has one
+  bedroom, one bathroom, one balcony") the word is counting. Items are
+  letter-led, so an enumeration over numbers is out; the distributive "one
+  for you, one for me, one for the pot" is skipped at any length. Nothing in
+  2.4M words of pre-2022 Hacker News or 1.25M words of public-domain prose.
+
+- `everyone-nobody` (structure, warning) flags the comma-spliced antithesis
+  on quantifier subjects: "Everyone wants the dashboard, nobody maintains
+  it." The
+  balance is what makes the diagnosis sound settled. The comma splice is
+  required (with "and" it is a sentence, with a period it is two), the two
+  subjects must differ in polarity, and the second clause must close the
+  sentence. Nothing in 2.4M words of pre-2022 Hacker News or 1.25M words of
+  public-domain prose.
+
+- `and-what-it-should` (rhetorical-tic, warning) flags the elliptical tail:
+  "List what the assistant knows about the client, and what it should." The
+  second clause borrows its verb from the first and closes on a bare modal
+  or a negated auxiliary, so the sentence ends on a contrast it never
+  states. The comma, the conjunction, and the full stop right after the
+  modal are all required; the affirmative copula and do-verb ("and what he
+  does.") are complete clauses and stay out, and so does a question.
+  Nothing in 2.4M words of pre-2022 Hacker News or 1.25M words of
+  public-domain prose.
+
+- `abstract-lives-in` (rhetorical-tic, info) flags an abstraction given an
+  address: "the craft that lives between the two desks", "its context lives
+  in a folder nobody else can open", "the value sits in the follow-up". The
+  subject list is closed and abstract, so people and dogs living and sitting
+  places never match, and a capitalised subject is a proper noun. "with" is
+  left out of the prepositions because "the decision sits with the board" is
+  ordinary English for who is responsible, "at" because "the value sits at
+  ten million" is a quantity, and "lies in" because "the problem lies in
+  the assumption" is where a fault is. It ships at info because the same
+  shape says where information literally is ("the knowledge lives in our
+  heads"), and a sample of pre-2022 Hacker News biased toward the
+  construction turns up a few of those per million words, all human. A
+  draft that keeps giving ideas addresses should be read as a warning.
+
+- `np-fragment-and` (structure, info) flags a whole sentence made of two
+  noun phrases and an "and": "A named owner and a quarterly review." It is
+  the fix half of a model's problem-then-fix pair with the verb left out,
+  and its usual habitat is a bulleted list, so a list marker may open it.
+  Each phrase is one to three words, and no auxiliary or modal may appear,
+  contractions included. It ships at info because a lexical verb is
+  invisible to the pattern: "A car and a truck collided." has the same
+  shape and flags. That sentence is rare in the corpora (nothing in 2.4M
+  words of pre-2022 Hacker News, once in 1.25M words of public-domain
+  prose), but it is a sentence, so one flag is a question; a draft full of
+  them should be read as a warning.
+
+- `the-x-is-the-x` (rhetorical-tic, warning) flags the repeated-head
+  equative: "the reason it holds up is the reason the other half happens",
+  "the problem with the tool is the problem with the team". A backreference
+  catches the same abstract head noun on both sides of the copula, so the
+  sentence equates two things while naming neither. The head list holds
+  only nouns that cannot name an object, since "the key to the front door
+  is the key on the red fob" is an identity statement; the two heads must
+  share a clause, so an earlier "the cost was low, but shipping is the
+  cost" never pairs; and the second head must be followed by a preposition,
+  determiner, quantifier, pronoun, plural noun or punctuation, so "the
+  answer key" is not a repeat. Once in 2.4M words of pre-2022 Hacker News;
+  nothing in 1.25M words of public-domain prose.
+
+- `same-determiner-chain` (rhetorical-tic, info) is the quiet cousin of
+  `one-x-one-y`: three or more comma-separated items opening on the same
+  determiner or quantifier, "every file, every branch, every deploy", caught
+  with a backreference. The narrative possessives (my, his, her, their, its)
+  are left out, since "his fame, his position, his life" is every
+  novelist's. It ships at info because the device is one humans use on
+  purpose: about 8 per million words of pre-2022 Hacker News and 14 per
+  million
+  in public-domain prose. One is a question; several in a draft should be
+  read as a warning.
+
+- `quip-question` (structure, info) flags the verbless question that opens
+  a pitch: "No invite?", "New to the tool?", "Still stuck?". It must start a
+  sentence, open on one of a short list of words, and close on the question
+  mark within four more words with no auxiliary or contraction, so a real
+  question stays out; "Need" and "Want" are left off the list because "Need
+  help?" is a question with its verb elided. Ships at info: about six per
+  million words of pre-2022 Hacker News, all of them replies asking the
+  same shape of a person; nothing in public-domain prose outside Joyce's
+  dialogue. Several in a draft should be read as a warning.
+
+- `is-the-whole-x` (rhetorical-tic, info) is `thats-the-whole` on any
+  subject: "That periodicity is the whole tell.", "Consistency is the real
+  test." It yields only the exact sentences the two older rules own
+  ("that/this is the whole point/game/thing/deal/story" and "is the entire
+  point/game/thing/deal/story"), so nothing is reported twice and "This is
+  the real test." is not lost. An interrogative subject is out, since a
+  question is not a closer; "only", "deal", "thing" and "cost" are left out
+  because "is the only thing", "the real deal", "the real thing" and "the
+  whole cost" are ordinary speech. Ships at info: about three per million words of
+  pre-2022 Hacker News, nothing in public-domain prose. Several in a draft
+  should be read as a warning.
+
+- `bare-equative` (rhetorical-tic, info) flags a sentence that opens on an
+  abstract head noun and equates it with a definite noun phrase: "The tell
+  here is the periodicity.", "The lesson is the handoff." The head list is
+  the one `the-x-is-the-x` uses, so "The key is the brass thing on the hook"
+  is a definition and stays out; the copula (including "isn't") must be
+  followed by "the" and a lowercase word, so "The problem is real", "The
+  answer is a mess", "the same", "the first" and "the Slack thread" are all
+  out; a list marker may open it. Ships at info: about four per million
+  words of pre-2022 Hacker News, nothing in public-domain prose. Several in
+  a draft should be read as a warning.
+
+- `mic-drop-closer` (structure, info) flags the kicker: a sentence of sixty
+  or more characters, then a two-to-eight-word closer that ends the
+  paragraph and opens on a quantifier or deictic, "Nothing here needs a new
+  login.", "Most teams end up with two." A blank line or the end of the
+  text must follow the closer, so a bullet followed by another bullet is
+  not one; both sentences may be hard-wrapped; and whitespace runs in the
+  long sentence are capped, so a URL blanked by `--markdown` cannot make it.
+  Ships at info because people end paragraphs this way too, at about 150
+  per million words of pre-2022 Hacker News (a comment ends on a verdict);
+  one flag means nothing. A draft where it repeats paragraph after
+  paragraph is the tell, and the rationale tells the agent to read that as
+  a warning.
+
+- `short-run` (structure, info) flags three consecutive sentences of thirty
+  characters or fewer, each closing on a full stop with no quotation mark:
+  "Nobody used it. A named owner. Then a review." The run must start at a
+  real sentence boundary, so the short tail of a hard-wrapped sentence never
+  opens one; dialogue is out by the boundary, sentences with digits as data,
+  initials and abbreviations as not sentence ends, consecutive bullets as a
+  list, and questions and exclamations by design. Ships at info: a staccato
+  run is a device people use on purpose, at about thirty per million words
+  of pre-2022 Hacker News. A draft that keeps doing it is the tell, and the
+  rationale tells the agent to read that as a warning.
+
+- `epistrophe` (rhetorical-tic, info, off by default) flags two clauses
+  that end on the same two-word phrase, the second closing the sentence:
+  "built for one desk, and almost no job is done at one desk." Two
+  backreferences catch the repeat, so the phrase may be hard-wrapped; an
+  article-led phrase, a short second word, a long or punctuated second
+  clause, and a clause made of blanked Markdown are all out. It is off by
+  default like `rule-of-three`: the figure is one Emerson and Marcus
+  Aurelius use on purpose, and on pre-2022 Hacker News most of the 42 hits
+  per million words are plain phrase reuse. Select it when a draft is
+  suspected of leaning on it; several then should be read as a warning.
+
+- `intersection-of` (rhetorical-tic, warning) flags "the intersection of X and
+  Y" used as positioning: a writer placed between two fields, saying nothing
+  about either. "at" is not required, so "explores the intersection of art and
+  technology" flags the same as "sits at the intersection of". Two guards keep
+  the literal senses out. A street corner names capitalised streets ("Elm",
+  "Broadway", "Highway 12"), so the word after "of" must be lowercase or an
+  all-caps acronym; "AI", "UX" and "HCI" are the metaphor's usual operands and
+  no street is spelled that way. Geometry and set arithmetic name their operands,
+  so a literal noun ("curves", "lines", "arrays", "ranges", "roads") within two
+  words of "of" drops the note.
+
+- `and-nothing-else` (rhetorical-tic, warning) flags a sentence that closes on
+  ", and nothing else", ", nothing more", ", nothing further" or ", and no
+  more". A model told to return one thing and nothing else carries the phrasing
+  into the prose it writes afterwards, where the exclusion repeats what the
+  sentence already said.
+- `nothing-else-frag` (rhetorical-tic, warning) flags the same exclusion built
+  as a fragment: "Return the JSON. Nothing else."
+- `honestly`, `honest-x` and `most-honest-x` (rhetorical-tic, warning) answer
+  the "honest" habit the way `cleanly`, `clean-x` and `cleanest-x` answer the
+  "clean" one. `honestly` takes the manner adverb closing on a full stop or
+  comma ("these two rows compare honestly"); `honest-x` takes the adjective in
+  front of the writer's own construction ("an honest comparison", "the honest
+  framing"); `most-honest-x` takes the self-ranking superlative.
+
+  `honestly` is anchored on position alone, so a dialogue tag ("said Isabel
+  honestly") and the sentence-final hedge of casual speech ("it's beyond boring
+  honestly") both match. Both are rare: twice in 3.65M words of public-domain
+  prose and ten times in 6.07M words of pre-2022 Hacker News. The two noun
+  rules flag nothing in either corpus; "an honest answer" and "an honest
+  assessment" are left out because on Hacker News they mean a person telling
+  the truth, not a writer praising their own framing.
+
+  Both rules are wide on purpose. There is no verb list and no imperative
+  requirement, so the only narrowing is structural, and English has always used
+  this tail: the pair flags three times in 1.92M words of public-domain prose
+  and twice in 461k words of pre-2022 Hacker News, and `and-nothing-else` flags
+  every refrain in "The Raven". Expect to dismiss it on fiction and on quoted
+  verse.
 
 ## [0.5.0] - 2026-08-27
 

data/README.md

@@ -10,7 +10,32 @@ A pattern earns a place in the catalog only if it shows up constantly in AI writ
 
 ## Installation
 
-Requires **Ruby 3.3+** and nothing else. The runtime is standard library only (`optparse`, `json`, native regex).
+Requires **Ruby 3.3+** and nothing else. The runtime is standard library only (`optparse`, `json`, native regex). The Ruby that macOS ships at `/usr/bin/ruby` is 2.6 and too old; `brew install ruby` gets you a current one.
+
+### In Claude Cowork or the Claude desktop app
+
+Open **Customize → Plugins → Add marketplace** and paste exactly this:
+
+```
+benjaminjackson/sloplint
+```
+
+Not the web address of the page you are reading — just those two words with the slash, and no slash on the end. `https://github.com/benjaminjackson/sloplint` works too, but a link copied from a page you were browsing usually has more on the end of it and will fail. Then install **sloplint** from the list.
+
+Installing this way needs no package install of any kind, which is why it works in Cowork, where `gem install` cannot reach the internet. Claude Code carries the source along with the plugin and runs it in place.
+
+New rules reach you when you click **Update** on the marketplace. Nothing updates on its own.
+
+Once it is installed, ask Claude to check a draft, or run `/sloplint:check`.
+
+### In Claude Code
+
+```
+/plugin marketplace add benjaminjackson/sloplint
+/plugin install sloplint@sloplint
+```
+
+### In a terminal
 
 ```bash
 gem install sloplint
@@ -35,7 +60,7 @@ The recipe sloplint is built around, and the one an agent should use:
 cat draft.md | sloplint check --markdown -o json -
 ```
 
-`--markdown` blanks out code and URLs first, `-o json` emits the machine-readable form, and `-` reads stdin. Exit 0 means clean, 1 means notes found, anything higher is an error. `check` is the default command, so `sloplint draft.md`, `sloplint -`, and a bare `sloplint` with piped stdin all scan.
+`--markdown` blanks out code, HTML comments, and URLs first, `-o json` emits the machine-readable form, and `-` reads stdin. Exit 0 means clean, 1 means notes found, anything higher is an error. `check` is the default command, so `sloplint draft.md`, `sloplint -`, and a bare `sloplint` with piped stdin all scan.
 
 The human-readable form drops `-o json`:
 
@@ -73,9 +98,10 @@ version      print the sloplint version
 
 `check` takes files as arguments, or `-` (or nothing) to read stdin, and these options:
 
-- `--markdown` skips fenced code, inline code, and URLs before scanning. Off by default so it never silently eats prose.
+- `--markdown` skips fenced code, inline code, HTML comments, and URLs before scanning. Off by default so it never silently eats prose.
 - `--select IDS` runs only these rules. Accepts comma-separated rule ids or category names.
 - `--ignore IDS` skips these rules. Same id-or-category form.
+- `--strict` runs every rule, including the five that are off by default. `--ignore` still applies on top.
 
 `explain` is the command an agent calls to decide whether a flag is worth acting on:
 
@@ -125,26 +151,28 @@ Three codes carry the contract. A crash exits nonzero on its own.
 | 1    | ran, notes found |
 | 2    | bad arguments or usage error |
 
-An unknown id or category in `--select`/`--ignore` is a usage error (exit 2, naming the id) rather than a silent no-op, so a typo can't masquerade as a clean scan.
+An unknown id or category in `--select`/`--ignore` is a usage error (exit 2, naming the id) rather than a silent no-op, so a typo can't masquerade as a clean scan. Input that is empty or only whitespace is exit 2 for the same reason: a pipe that delivered nothing must not read as a clean draft. Only when every source is empty — one empty file among several named ones is taken as deliberate.
 
 ## The rule catalog
 
-50 rules across four categories. `sloplint rules` prints them; `sloplint rules --json` gives an agent the enumerable form.
+80 rules across four categories. `sloplint rules` prints them; `sloplint rules --json` gives an agent the enumerable form.
 
-- **rhetorical-tic** (36) the cadence patterns: `no-x-no-y`, `no-x-no-y-frag`, `thats-the-whole`, `thats-how-x`, `announced-takeaway`, `exact-exactly`, `load-bearing`, `you-already-know`, `sit-with-that`, `hold-onto-that`, `cleanly`, `clean-count`, `cleanest-x`, `clean-x`, `not-nothing`, `is-is` (doubled copula), `if-im-being-honest` (the candor preamble, from slopwash.com's "false intimacy"), and more.
-- **puffery** (5) Wikipedia's "signs of AI writing": `puffery-words` (vibrant, nestled, groundbreaking, in the heart of), `rich-tapestry`, `vital-role`, `stands-serves-as`, `underscores-highlights`.
-- **structure** (8) `not-just-x-but-y`, `not-x-but-y` (the bare corrective), `em-dash` (any em dash), `em-dash-overuse` (three or more in one paragraph), `question-isnt` (the corrective frame in interrogative dress), `less-about-more-about`, `trailing-significance-participle` (the "…, showcasing its importance" clause), and `rule-of-three`.
+- **rhetorical-tic** (56) the cadence patterns: `no-x-no-y`, `no-x-no-y-frag`, `from-x-to-y-chain` ("from guessing to measuring, from hoping to knowing"), `one-x-one-y` ("one reviewer, one queue, one deadline"), `same-determiner-chain` (any other repeated determiner, at `info`), `and-what-it-should` (the elliptical tail: "…, and what it should."), `abstract-lives-in` ("the value sits in the follow-up", at `info`), `the-x-is-the-x` ("the problem with A is the problem with B"), `bare-equative` ("The lesson is the handoff.", at `info`), `thats-the-whole`, `is-the-whole-x` (the same closer on any subject: "Consistency is the real test.", at `info`), `thats-how-x`, `announced-takeaway`, `exact-exactly`, `load-bearing`, `intersection-of`, `you-already-know`, `sit-with-that`, `hold-onto-that`, `cleanly`, `clean-count`, `cleanest-x`, `clean-x`, `not-nothing`, `is-is` (doubled copula), `if-im-being-honest` (the candor preamble, from slopwash.com's "false intimacy"), `and-nothing-else` (the trailing "…, and nothing else"), `honestly`, `honest-x`, `most-honest-x` (the honesty family, built the same way as the four `clean` rules above), `impact-verb` ("the outage impacted four thousand accounts"), `impact-noun-bare` ("the impact of X", at `info` because research prose uses it straight), and more.
+- **puffery** (6) five from Wikipedia's "signs of AI writing" — `puffery-words` (vibrant, nestled, groundbreaking, in the heart of), `rich-tapestry`, `vital-role`, `stands-serves-as`, `underscores-highlights` — plus `impact-noun-vague` ("a significant impact", "make an impact").
+- **structure** (17) `everyone-nobody` (the comma-spliced antithesis: "Everyone wants the dashboard, nobody maintains it."), `np-fragment-and` (the verbless "A named owner and a quarterly review.", at `info`), `quip-question` (the verbless "No invite?", at `info`), `mic-drop-closer` (the short quantifier-led sentence that ends a paragraph after a long one, at `info`; a draft full of them is a warning), `ellipsis-closer` (the same shape, but the closer's verb is elided down to a bare auxiliary: "The agent did.", at `info`), `short-run` (three sentences of thirty characters or fewer in a row, at `info`, same caveat), `not-just-x-but-y`, `not-x-but-y` (the bare corrective), `isnt-x-its-y` (the same corrective split across two clauses: "It isn't the tool. It's the habit."), `not-by-x-but-by-y` (the corrective on a repeated preposition: "not by luck, but by design"), `em-dash` (any em dash), `em-dash-overuse` (three or more in one paragraph), `question-isnt` (the corrective frame in interrogative dress), `less-about-more-about`, `trailing-significance-participle` (the "…, showcasing its importance" clause), `trailing-restatement` (the "…, which means …" tail that says the sentence again, off by default), and `rule-of-three`.
 - **hedging** (1) `vague-attribution`: "some critics argue," "it is widely regarded."
 
 Severity is `warning` for strong tells, `info` for weak or contextual ones. No rule currently ships at `error`; the tier is reserved for a pattern with essentially zero false-positive risk, and none has earned that yet.
 
 Some tells come in a confident form and an ambiguous one, and those ship as a pair rather than as one rule stretched over both. `no-x-no-y` wants the comma chain a writer clearly authored; `no-x-no-y-frag` takes the same cadence built from sentence fragments, which ordinary prose also produces, and ships at `info`. Same with `not-just-x-but-y` and `not-x-but-y`, and with `notice-what-there` and `notice-what`. The quiet half is still worth flagging — an agent that reads the rationale can judge — but it should not carry the same weight as the half we're sure about.
 
-Two rules ship **off by default**, and both run only when you name them: `sloplint check --select rule-of-three -`. `rule-of-three` flags three parallel comma items closing a sentence, which humans do all the time. `genuinely` flags every occurrence of the word; as an intensifier it rates the writer's sincerity, but it still does real work when it draws a contrast, and nothing in the sentence separates the two.
+`and-nothing-else` and `nothing-else-frag` are a pair of the same shape, but both ship at `warning`. That is a deliberate exception: the fragment half carries a capital letter and a whole-sentence requirement that the comma half has no equivalent of, so it is the *narrower* of the two rather than the quieter one.
+
+Five rules ship **off by default**. They run when you name them — `sloplint check --select rule-of-three -` — or when you pass `--strict`, which turns the whole catalog on. `rule-of-three` flags three single words in a comma series closing a sentence, which humans do all the time; the closing two items must be single words, so a triad of phrases does not match, because a regex cannot tell one from an ordinary list. `genuinely` flags every occurrence of the word; as an intensifier it rates the writer's sincerity, but it still does real work when it draws a contrast, and nothing in the sentence separates the two. `epistrophe` flags two clauses ending on the same phrase, a named figure that careful writers use on purpose and that, on Hacker News, is mostly plain phrase reuse. `trailing-restatement` flags the "…, which means …" tail and the participles that hang a result off the sentence ("…, making it easier"); the connective is visible and the restatement is not, so a real consequence flags the same way. `phrase-echo` flags a three-word phrase that comes back within a few hundred words; a term of art comes back because it must, and the pattern cannot tell one from a phrase the writer coined.
 
 ### Markdown handling
 
-`--markdown` replaces fenced code, inline code, and URLs with same-length whitespace before scanning, so line and column stay correct. Without it, sloplint treats the whole file as prose and will flag text inside your code fences. Pass `--markdown` whenever the input is Markdown.
+`--markdown` replaces fenced code, inline code, HTML comments, and URLs with same-length whitespace before scanning, so line and column stay correct. Without it, sloplint treats the whole file as prose and will flag text inside your code fences. Pass `--markdown` whenever the input is Markdown.
 
 ## Adding a rule
 

data/docs/SPEC.md

@@ -93,13 +93,15 @@ dependencies at runtime.
 Rationale: rules are regexes; the whole thing is a scanner plus an output
 formatter. A dependency-free `gem install sloplint` is the robust, boring
 choice. The name is free on RubyGems (taken on PyPI, so this also sidesteps the
-collision). Ships a `sloplint` executable via the gemspec's `bin`.
+collision). Ships a `sloplint` executable from `exe/` (claude.ai rejects a
+plugin with a top-level `bin/`, and sloplint ships as a Claude Code plugin
+too).
 
 Package layout (standard gem):
 
 ```
 sloplint/
-  bin/sloplint             # thin shim: require "sloplint/cli"; exit Sloplint::CLI.run(ARGV)
+  exe/sloplint             # thin shim: require "sloplint/cli"; exit Sloplint::CLI.run(ARGV)
   lib/sloplint.rb          # requires the pieces below
   lib/sloplint/version.rb
   lib/sloplint/cli.rb      # optparse, subcommands, exit codes
@@ -136,9 +138,10 @@ global options:
 
 check options:
   paths ...            files to scan; "-" or no paths reads stdin
-  --markdown           skip fenced/inline code spans
+  --markdown           skip fenced/inline code spans, HTML comments, and URLs
   --select IDS         only run these rules (comma-separated ids or categories)
   --ignore IDS         skip these rules
+  --strict             run every rule, including the off-by-default ones
 ```
 
 `check` is the default command. A first argument that is not a command name is
@@ -161,6 +164,11 @@ Three codes carry the contract. A crash just exits nonzero on its own.
 | 1    | ran, **notes found** |
 | 2    | bad arguments / usage error |
 
+Empty or whitespace-only input is exit 2, like a mistyped rule id: a scan of
+nothing must not report as a clean scan. The text is tested before
+`--markdown` blanks code, HTML comments, and URLs, so a file that holds only a fenced code
+block still exits 0. Only when every source is empty.
+
 ## Note (the diagnostic object)
 
 One match = one Note. JSON output is an array of these (or an object keyed by
@@ -255,7 +263,7 @@ Categories (for `--select`/`--ignore` by group):
 
 - `rhetorical-tic` — the cadence patterns (the user's list below)
 - `puffery` — Wikipedia "words to watch" (boasts, vibrant, nestled, tapestry…)
-- `structure` — rule-of-three, "not just X but Y", "the question isn't X, it's Y", "less about X more about Y", the trailing significance participle, em dash, em-dash overuse
+- `structure` — rule-of-three, "not just X but Y", "the question isn't X, it's Y", "less about X more about Y", the trailing significance participle, the trailing restatement, em dash, em-dash overuse
 - `hedging` — vague attribution ("some critics argue", "it is widely regarded")
 
 Severities: `warning` for strong tells, `info` for weak/contextual ones. No
@@ -270,12 +278,20 @@ false-positive risk, which none has demonstrated.
 |----|---------|-------|
 | `no-x-no-y` | 2+ comma-separated "no …" items in a row | counts items |
 | `no-x-no-y-frag` | the same cadence as sentence fragments ("No fluff. No filler.") | counts items; `info` |
-| `thats-the-whole` | "that/this is the whole point/game/thing…" | |
+| `thats-the-whole` | "that/this is the whole point/game/thing…" | also value and fix, the closers agents write in technical prose, when they end the sentence or the paragraph or run into a closer's tail word; either apostrophe; `is-the-whole-x` yields the same list |
+| `is-the-whole-x` | any subject + "is the whole/real/actual/entire N" (tell, point, test, work, …) | opens on the subject word; yields only the exact sentences `thats-the-whole` and `is-the-entire` own; interrogative subjects out; "only", "deal", "thing", "cost" left out; `info` |
 | `did-not-x-did-not-y` | 2+ "did not …"/"didn't …" in a row | counts items |
+| `from-x-to-y-chain` | 2+ comma-separated "from X to Y" spans | counts spans; operands open with a letter, so ranges are out; skips the relay ("to B, from B") and the reduplication ("from X to X") |
+| `one-x-one-y` | 3+ comma-separated "one X" items standing on their own | counts items; must open a sentence or follow a colon, so a chain after a verb is counting; items letter-led; the distributive "one for …" skipped; a pair never flags |
+| `and-what-it-should` | ", and what it should." — a second "what" clause closing on a bare modal or a negated auxiliary | comma, conjunction and full stop required; the affirmative copula and do-verb ("what he does.") are complete clauses and out; a question is out |
+| `abstract-lives-in` | an abstract noun that "lives/sits in/between/inside/…" | closed subject list; capitalised subjects skipped; "with" (responsibility), "at" (quantity), "lies in" and "resides in" left out; `info` |
+| `the-x-is-the-x` | "the X … is the X …": the same abstract head noun on both sides of the copula | backreference; closed list of heads that cannot name an object (key, cost, unit are out); the clause between is capped at 50 chars and may not hold a comma, semicolon or colon; the second head must be followed by a preposition, determiner, quantifier, pronoun, plural noun or punctuation, so compounds are out; "isn't" counts |
+| `same-determiner-chain` | 3+ comma-separated items opening on the same determiner or quantifier (every, each, your, more, …) | backreference; counts items; "one" and "no" left to their own rules; the narrative possessives (my, his, her, their, its) left out; items lowercase-led, so proper nouns are not a chain; `info` |
+| `bare-equative` | sentence-initial "The N (here) is (not) the …" with an abstract head noun | the-x-is-the-x's head list, so concrete heads are out; the copula (is, is not, isn't) must be followed by "the" and a lowercase word, so predicate adjectives, indefinites, pointing complements ("the same/one/first/…") and proper nouns are out; a list marker may open it; `info` |
 | `dont-verb-it` | "Don't call it X. Call it Y." (negated verb+it, same verb+it) | |
 | `sit-with-that` | "sit with that/this/it", "sit with the discomfort" | |
 | `hold-onto-that` | sentence-initial "hold onto/on to that/this" | imperative only |
-| `cleanly` | "cleanly" anywhere | no verb list; the engineering idiom counts too |
+| `cleanly` | "cleanly" into/onto/in two/in half — the partition frame | the preposition is the narrowing; the engineering idiom ("applies cleanly", "separated cleanly", "cleanly compiled") is a checkable fact and stays out, and the clause-final form is left out. `info`: the frame is not the sense, and separating "splits cleanly into two parts" from "retracted cleanly into the well" needs the subject |
 | `clean-count` | "two/three clean parts/buckets/categories…" | needs a partition noun |
 | `cleanest-x` | "the cleanest framing/formulation", "cleanest way to put it" | noun list only |
 | `clean-x` | "a clean abstraction/distinction/framing", "clean line between" | `info` |
@@ -283,12 +299,16 @@ false-positive risk, which none has demonstrated.
 | `is-the-entire` | "X is the entire point/game/business model" | |
 | `the-entire-is` | "the entire point/game/… is" (flip of above) | |
 | `is-real-and-not` | "the X is real, and/not…", "is the real … and it" | skip "real estate/time"; `info` |
+| `real-x-real-y` | the same "real" used twice attributively in one sentence, in front of two different nouns: "real API calls … real credentials" | needs no noun list -- narrowed by requiring two attributive uses of the intensifier naming two different things; a hyphen on either side of "real" (real-time, non-real) takes it out of the running; a closed list drops the fixed senses (real time, real-world, real numbers/roots, real user (monitoring), real estate, real money) and the function words that continue a predicative "is real" |
 | `the-punchline-is` | "the punchline is/:/?", "the honest answer/version is" | "short version" left out; ordinary writing |
 | `worth-naming` | "worth naming/flagging/separating/spelling out" | skip "naming names"; yields to the rule below when a manner adverb follows; `info` |
 | `worth-saying-plainly` | "it's worth saying plainly / better put bluntly…", plus the bare "Put plainly," / "Said bluntly," | sentence-initial; the bare branch drops "simply"/"clearly" so "put simply" and "simply put" stay clean |
 | `not-nothing` | copula + "not nothing" litotes, any subject | skip personal/there subjects |
-| `exact-exactly` | "exact"/"exactly" | allowlist for the checkable uses; `info` |
+| `exact-exactly` | the reflexive intensifier: "that's exactly", "exactly right", "exactly the point", "the exact problem", "know exactly why" | matches the tell as a closed set of frames rather than matching the word and subtracting an allow-list; a measured quantity ("the exact diameter", "the exact CPU time") is silent by construction. `info` |
 | `load-bearing` | "load-bearing" outside its construction sense | skip building nouns either side |
+| `intersection-of` | "the intersection of X and Y" as positioning | skip street corners, geometry, set arithmetic, airfield surfaces (runway, taxiway, apron), matrix rows and columns, and operands shaped like a US route designator ("US-27A") or a quadrant plus house number ("NE 140th Court") |
+| `impact-verb` | "impact" used as a verb ("the outage impacted 4,000 accounts") | needs an auxiliary or subject pronoun for the base form; "to impact" requires a following object, so the preposition ("prior to impact") stays out; skip the medical and soil sense of "impacted", the struck object of a real collision ("impacted terrain"), the fixed compounds, and hyphenated forms |
+| `impact-noun-bare` | "the impact of X", "measure the impact" — `info` | needs a measuring verb in front or "of" behind; skip the collision sense and the fixed compounds |
 | `thats-how-x` | sentence-initial "that's how…" | |
 | `announced-takeaway` | colon-led label: "The pattern/lesson/takeaway…:" | sentence-initial |
 | `earns-its-place` | "earns its place/keep" (any possessive) | possessive required; `warning` |
@@ -301,8 +321,16 @@ false-positive risk, which none has demonstrated.
 | `notice-what` | bare sentence-initial "Notice what…" | yields the "there" frame to the rule above; "how" excluded; `info` |
 | `none-of-this-is-to-say` | "none of this/that/the above is to say" | every other "not to say" phrasing excluded |
 | `if-im-being-honest` | "if I'm/we're (being) honest", "honestly, the answer/truth" | plain "to be honest" and "I'll be honest" excluded |
+| `honestly` | a word, then "honestly", then a full stop or comma | terminal position required; discourse-marker slot guarded ("and/quite/but honestly,"); no animacy test |
+| `honest-x` | "an honest comparison", "the honest framing" | short noun list; "answer", "assessment", "account" excluded as ordinary; superlative yielded to the rule below |
+| `most-honest-x` | "the most honest framing", "the most honest way to put it" | wider noun list than `honest-x`; no human nouns, so "the most honest person" stays out |
 | `genuinely` | any "genuinely" | off by default; no narrowing holds |
+| `actually-not-x` | "actually" and a trailing "…, not X" in one clause | the comma before "not" must be the first comma of the clause and must follow a word, so a fronted setup ("Despite the name, …") and a parenthetical both drop it; never crosses a line break; bare "actually" and "not actually" both left out |
+| `epistrophe` | two clauses ending on the same two-word phrase, the second closing the sentence | two backreferences, so the phrase may be hard-wrapped; no article-led phrase, second word 4+ letters, second clause 5–60 chars with no internal punctuation and capped whitespace; off by default; `info` |
+| `phrase-echo` | the same three words again within about 400 words | backreference in a lookahead, so the match is the first occurrence; each word 4+ characters and lowercase-led, one of them 6+ letters with nothing but letters, so function-word runs, contractions, proper nouns and Title Case headings are out; a repeat that opens on a quote mark, backtick, emphasis marker or hyphen is out; the gap crosses paragraph breaks but not a list item, a table row or 80+ non-word characters; off by default; `info` |
 | `and-thats-fine` | "And that's fine/okay." as a whole sentence | "and" required; must open and close the sentence |
+| `and-nothing-else` | trailing ", and nothing else/more/further", ", and no more" | tail must close the sentence; comma required; bare "no more" needs "and"; "?" excluded |
+| `nothing-else-frag` | the same exclusion as a fragment: "Nothing else." | sentence-initial capital; semicolon excluded; must be the whole sentence; "No more." left out |
 | `is-is` | doubled copula: "what it is is …", "the thing is, is that …" | comma optional |
 
 ### puffery (Wikipedia: Signs of AI writing)
@@ -314,15 +342,25 @@ Single flat rule per word-cluster, matched as whole words:
 - `vital-role` — "plays a (vital/crucial/pivotal/significant/key) role".
 - `underscores-highlights` — "underscore(s)" + determiner and "underscored/underscoring" anywhere (the emphasis verb); "highlights/emphasizes its (importance/significance)" stays narrow.
 - `rich-tapestry` — "rich tapestry", "tapestry of".
+- `impact-noun-vague` — an intensity adjective plus "impact" (significant, real, meaningful, lasting, massive, huge, profound, big and the rest), or "make/have an impact". "positive" and "negative" stay out: they name a direction, which is more than the intensity words do. "statistically significant impact" is skipped — that one is a finding.
 
 ### structure
 
 - `not-just-x-but-y` — copula + "not just/only/merely/simply/solely X … but (also) Y", plus "not because X, but because Y". Requires the escalation word.
 - `not-x-but-y` — the bare corrective "is not X but Y" with no escalation word; `info`, because the corrective/concession distinction is syntactic and the pattern can only approximate it.
+- `isnt-x-its-y` — the corrective with no conjunction: a negated copula, then a second clause that supplies the replacement ("It isn't the tool. It's the habit."), joined by a period, semicolon, comma, or dash; `info`. Neither complement may open with a pronoun, possessive, preposition ("about" excepted), degree word, or predicate adjective, which keeps out ordinary two-part contrast; some hits are still the same shape written by a person.
+- `not-by-x-but-by-y` — the corrective anchored on a repeated preposition rather than a copula: "not by A, but by B", "not from A but from B" (by, for, from, with, about, in, on, at, to, of, through, because of, out of); `info`. The B side must repeat the A preposition, A is capped at six words, and B may not open with a pronoun. Every corpus hit is the real shape written by a person, which is why it sits at info rather than being narrowed further.
+- `everyone-nobody` — the comma-spliced antithesis on quantifier subjects: one clause opens on everyone/everybody, the other on nobody/no one/none or "one N", joined by a bare comma, the second closing the sentence ("Everyone wants the dashboard, nobody maintains it."). A conjunction, a period, or the same subject twice is not a hinge; "one of/by/per", a measure, a proper noun, "none of which" and "none louder than" are not second subjects; clauses are capped at eighty characters.
+- `np-fragment-and` — a whole sentence made of two noun phrases and an "and", opening on A/An/One at a sentence start or after a list marker ("A named owner and a quarterly review."). One to three words a side, no auxiliary or modal anywhere (contractions included); a lexical verb is invisible, so "A car and a truck collided." flags, which is why it ships at `info`.
+- `quip-question` — the verbless opening question ("No invite?", "New to the tool?", "Still stuck?"): sentence-initial, one of a short list of opening words, one to four more words, no auxiliary or contraction, closing on the question mark; "Need" and "Want" are left out as elided verbs; `info`, because people ask the same shape in conversation.
+- `mic-drop-closer` — a sentence of 60+ characters, then a paragraph-final closer of two to eight words opening on a **quantifier** (Nothing, Most, None, Everything, Everyone, Nobody, Then, Neither, Both): "Nothing here needs a new login."; `info`. The bare demonstratives (That, This, It) were in the list and are out: procedural writing ends a step with one as a matter of course ("This completes the roughing operations."). A blank line or the end of the text must follow the closer; both sentences may be hard-wrapped; whitespace runs in the long sentence are capped so blanked Markdown cannot make it, and the long-sentence prefix is an atomic group so an unpunctuated stretch cannot send it into catastrophic backtracking. The note points at the closer. One means nothing; a draft where it repeats is the tell.
+- `ellipsis-closer` — the same long-sentence-then-short-closer shape as `mic-drop-closer`, but the tell sits in the verb rather than the subject: the closer's verb phrase is elided down to a bare auxiliary with no object, "The agent did."; `info`. No subject list is needed — a one-to-three word subject runs straight into a bare `did/does/do/was/were/is/are/had/has/can/could/would/will/should/might/must` (contracted forms included) and a period, and the closer must be the last thing in the paragraph, same as `mic-drop-closer`, so a closing quotation mark after the period excludes quoted dialogue the same way `short-run` excludes it. A negative lookahead drops a closer that still holds "what", "that", "which", "who", "why" or "how", since those introduce a subordinate clause supplying its own complement rather than an elided one ("Nobody knew who did." asks who did it). Reuses `mic-drop-closer`'s long-sentence prefix rather than a second copy of it. One means nothing; a draft where it repeats is the tell.
+- `short-run` — three consecutive sentences of thirty characters or fewer, each letter-led and closing on a full stop, no quotation marks or digits, no lone-letter labels (either case, though a possessive is not one) or abbreviations, starting at a real sentence boundary (never on a wrap continuation), crossing a hard wrap but not a paragraph break; `info`. A list marker may open a run but never sit inside one, so consecutive bullets are a list — the marker set covers glyphs, the literal "o" plain-text documents use as a bullet, and numbered and lettered items. Two whole-run exclusions keep document furniture out, each a property of the run rather than of any sentence in it: three single-word sentences in a row is a citation line ("Natl. Inst. Stand. Technol."), and three lowercase-led sentences in a row is transcribed speech ("we all get gas. we go to divert to Albany."). One single-word sentence is the archetypal kicker and stays, and a run that reaches a capital anywhere is prose, so identifier-initial writing ("npm was slow. git blame helped. We moved on.") is untouched. One is a question; a draft that repeats it is the tell.
 - `question-isnt` — "the question isn't/is not (whether|if|how|what|why|who) X, it's/but Y"; `info`. The resolving clause is required, so a plain rhetorical question never matches; "the real question is" is excluded.
 - `less-about-more-about` — "it's/this is/that's less about X (and) more about Y", also "… than about Y"; `info`. Both halves of the frame are required, and the subject slot is limited to the pronouns.
 - `trailing-significance-participle` — comma plus a participle from a closed verb list (highlighting, showcasing, reinforcing, shaping, enhancing, cementing, solidifying, embodying, fostering, facilitating, signalling), the clause a model hangs off a sentence to say what a fact means. Guards drop gerund lists and "signalling to". `driving`, `representing`, `reflecting`, `marking`, `contributing`, `illustrating`, `demonstrating`, `emphasising`, `echoing` and `affirming` stay out: humans write them in the same position, usually with a person as the subject, and the pattern cannot see the subject. `underscoring` is left to `underscores-highlights`.
-- `rule-of-three` — three parallel comma items ending a sentence (heuristic; `info` severity, off by default via `--select` since it false-positives).
+- `trailing-restatement` — comma plus "which means", "which is to say", or "meaning" opening on one of a closed set of determiners and pronouns; or one of four participle frames with a pronoun object and a closing word ("making it easier", "allowing us to", "giving them more", "leaving you with"). The closer on "making" is a comparative that ends the clause or leads into "to", "for" or "than". Nothing before the comma is inspected, so a gloss ("_ma_, which means hand") matches; the bare participles ("leaving the door open") never do. Off by default: the pattern sees the connective, not whether the tail restates the head. `info`.
+- `rule-of-three` — three single-word comma items ending a sentence (heuristic; `info` severity, off by default; runs under `--select` or `--strict` since it false-positives).
 - `em-dash` — any em dash; `info`.
 - `em-dash-overuse` — 3+ em dashes in one paragraph; `warning`.
 
@@ -332,8 +370,8 @@ Single flat rule per word-cluster, matched as whole words:
 
 ## Markdown handling
 
-`--markdown` blanks out fenced code (```` ``` ````), inline code (`` ` ``), and
-URLs before scanning, replacing them with same-length whitespace so line/column
+`--markdown` blanks out fenced code (```` ``` ````), inline code (`` ` ``), HTML
+comments (`<!-- -->`), and URLs before scanning, replacing them with same-length whitespace so line/column
 stay correct. Off by default (plain-text mode) so it never silently eats prose.
 
 ## Agent-first help text