no-ai-slop

Hard-rule audit that strips AI slop: emdashes, intensifiers, filler, hollow claims, fabricated facts, AI verbs and transitions, dramatic headings, weasel words — 24 non-negotiable rules with WRONG/RIGHT examples and a banned-words reference. Complementary to avoid-ai-writing (tiered detector) and the Biber register locks.

Source: .opencode/skills/no-ai-slop/SKILL.md — site rebuilt 2026-09-05.

No AI Slop — Hard Rules with OKF Adaptation¶

Upstream: realrossmanngroup/no_ai_slop_writing_rules (MIT). This skill is a port — the 24 rules, WRONG/RIGHT examples, and references/ai-writing-detection.md are kept verbatim. The section immediately below adapts them to legal drafting at The Building Law Practice. Where a note and the catalog conflict, the note wins only for the OKF case it names.

OKF workspace notes (BLP adaptation)¶

This is a port of the Rossmann Group no-ai-slop rules v1.0.0 (MIT). The catalog below is upstream verbatim; these notes adapt it to Scots legal drafting without weakening any rule.

Where it sits in the drafting pipeline¶

register lock (deepseek-flash-biber-fact / deepseek-flash-biber-argument
  and the other 17 Biber locks — fact vs argument, never mixed)
    → no-ai-slop (this skill — 24 hard rules, banned-words self-check)
    → avoid-ai-writing (tiered detector, 0-100 score, voice options)
    → sub-editor (grammar/punctuation pass — Swan/Seely/Cutts; PASS/FAIL gate)
    → court-auditor (court documents only — .opencode/house-style/court.md; mandatory PASS before shown)

This skill never replaces a gate. Its output is a draft like any other; sub-editor and court-auditor still run over it. The full 24-rule list lives in CLAUDE.md in this folder (copied from upstream CLAUDE.md); the actionable subset with WRONG/RIGHT examples is in the catalog below.
Complements avoid-ai-writing, does not duplicate it. no-ai-slop is binary hard rules (violation = unusable). avoid-ai-writing is a tiered detector (Tier 1A always-flag, Tier 2 cluster, Tier 3 density) with a deterministic 0-100 score via detector/patterns.js in that skill. Run both: hard rules first, then the tiered scan for residual tells.
Citation discipline is shared. Rules 2, 5, 17, 18, 19 (no unsourced statistics, no hollow statements, no fabricated scenarios/history/attributions) and Rule 23 (quote accurately) map directly to OKF's case-law and okf-ocr contracts: every factual proposition carries its Inventory of Productions citation (e.g. "Third Inventory, pp. 724–779"), every case citation carries a confirmed paragraph pinpoint + verbatim chunk, every document figure is verbatim. The fact-checker agent gates this afterwards — this skill flags, it never invents a citation to fill a gap.
Research-process narration (Rule 24). Report facts you can support and silently omit what you cannot. In OKF matters, residual gaps are marked [CONFLICT] / [LOW CONFIDENCE] per okf-ocr and verbatim-ocr, not narrated as "could not be located" or "as of [date]". The self-check must delete unsupportable sentences rather than hedge them.

Register precedence (Biber locks outrank style convenience)¶

The Biber register locks set Dimension 1 targets this skill must not flatten:

Nominalized abstractions, suasive modals (must, should, ought, require, entitle), and agentless passives where the agent is the law or the court ("it was held", "decree should be pronounced") are deliberate in the argument register (deepseek-flash-biber-argument and siblings). Do not cut them as "weasel words" or "nominalization slop".
Past-tense third-person narration with no evaluation is deliberate in the fact register (deepseek-flash-biber-fact and siblings). Do not inject present-tense suasion into a fact paragraph to satisfy Rule 13's "researcher" impulse.
If a rewrite would remove a register-required feature to satisfy an anti-slop rule, keep the register feature and note it as intentional. This skill is an AI-tell clean, not a plain-English conversion — that is bingham-plain and cutts-plain-english, which are never mixed with Biber.

Pleading-convention carve-outs (Scots court work)¶

Standard Scots pleading connectors and formulas are house convention, not AI tells. In court documents, keep them and in detect mode label them "convention, not AI":

"Further", "Separately", "In any event", "In these circumstances", "The pursuer avers", "It is submitted that", "for the reasons set forth above", "the said works", "esto", "quantum", credential enumerations.
The same holds for legacy phrases inside quoted or attributed material — the case-law flag-don't-fix rule already protects quotes, code blocks, tables, and attributed text.
Headings (Rule 16 vs house style). For court documents, .opencode/house-style/court.md governs: Palatino Linotype, CRAVES / CONDESCENDENCE / PLEAS-IN-LAW in caps centred bold 12pt, numbered paragraphs with hanging 1.5 cm tabs. Do not rewrite those to "descriptive" headings. court-auditor remains the final arbiter. For non-court prose, Rule 16 applies verbatim: headings must be concrete and descriptive, never thriller-style ("The Hidden Cost…", "The Trap").
Typography. The banned emdash (Rule 1) aligns with OKF house style: straight quotes, no serial comma, sparing em dashes. Target zero emdashes per 1,000 words. In court documents, use commas, periods, or parentheses instead. The --style okf-court config in avoid-ai-writing/examples/okf-court.json encodes the same preference.

How to use this skill¶

Before writing: read this SKILL.md and references/ai-writing-detection.md (the full banned-verbs/adjectives/transitions/hedging/heading-anti-pattern lists).
After drafting: self-check against references/ai-writing-detection.md and the 10-step checklist at the end of the catalog. Fix what you find. Then run avoid-ai-writing for the tiered 0-100 scan if a score is wanted.
Court documents: draft in the correct Biber register, then run this skill with the carve-outs above, then sub-editor, then court-auditor. Nothing is shown with a FAIL.

What this skill does not do¶

It does not set Biber scores — that is bieber-scale.
It does not teach Scots pleading doctrine — that is pleading (Black/Macphail).
It does not invent authorities — that is forbidden everywhere in this workspace.

Catalog (upstream verbatim)¶

The full rule list lives in CLAUDE.md (rules 1 through 24) in this folder. This section turns the rules that have worked examples into actionable guidance: each shows a WRONG version (the slop) and a RIGHT version (the fix). The pattern behind every fix is the same: replace the vague claim with a specific, checkable fact.

Rule 1: No emdashes¶

The character is banned. Use a semicolon, a period, a comma, or restructure.

WRONG: "The policy -- which affected millions -- was later reversed."
RIGHT: "The policy affected millions of devices. The company reversed it in December 2017."

Rule 4: No intensifiers¶

"Significantly", "dramatically", "extremely" and their kin are placeholders for evidence. Replace the word with the number it was standing in for.

WRONG: "The pricing was significantly higher than the cost of the part."
RIGHT: "They charged $1,200 for a repair that needed a $5 chip."

Rule 5: No hollow statements¶

A sentence that asserts importance without a detail says nothing. End every claim on a concrete fact.

WRONG: "This practice has had a significant impact on people."
RIGHT: "The company replaced 11 million batteries in 2018, against the 1 to 2 million it had expected."

Rule 7: No structural slop (repetitive layouts)¶

Three sections built from the same template read as machine output, even when each fact is true. Vary paragraph count, sentence rhythm, and how each section opens.

WRONG (three sections, identical shape):

In [year], [party] did [thing]. This affected [number] people. [Party] responded by [action].
In [year], [party] did [thing]. This affected [number] people. [Party] responded by [action].
In [year], [party] did [thing]. This affected [number] people. [Party] responded by [action].

RIGHT (vary the shape):

Section one: a detailed narrative with timeline and context across two paragraphs.
Section two: a two-sentence summary, because the event is thinly documented.
Section three: opens with the party's stated justification, then the contradicting evidence.

Rule 11: No filler phrases¶

"In today's world", "It's important to note", "When it comes to" add length, not meaning. Open on the fact.

WRONG: "In today's world, planned obsolescence affects many devices."
RIGHT: "Apple, Samsung, and Google have each faced lawsuits alleging planned obsolescence."

Rule 13: Write like a researcher, not a copywriter¶

If a sentence could sit on any advocacy or marketing site without changing a word, it is generic. Anchor it to something checkable.

WRONG: "People deserve the right to repair their own devices."
RIGHT: "The FTC voted 5-0 in July 2021 to step up enforcement against illegal repair restrictions."

Rule 15: No weasel words¶

"May potentially", "can help to", "might be able to" hedge a claim into meaninglessness. Either the thing happens or it does not. Say which.

WRONG: "Serialization may potentially prevent independent repair."
RIGHT: "Replacing an iPhone 15 camera module without the manufacturer's calibration software disables optical image stabilization."

Rule 16: No dramatic headings¶

A heading names what the section holds. It does not tease, dramatize, or abstract.

WRONG: "The Hidden Cost of Planned Obsolescence"
RIGHT: "Economic impact of shortened product lifespans"

Rule 19: No fabricated attributions¶

Never put a position in a named person's mouth from inference. State only what they actually did or said, with the real source.

WRONG: "Senator Smith has argued that the right to repair is essential."
RIGHT: "Senator Smith co-sponsored the Fair Repair Act in January 2024."

Root-cause differentiation¶

When you contrast two things, name the concrete difference that separates them. Do not assert that one is exempt, newer, better, or unaffected without saying what specifically makes it so.

WRONG: "2020+ Leaf models are unaffected and use the MyNISSAN app instead."
RIGHT: "2020+ Leaf models shipped with 4G/LTE telematics units connected to a newer cloud platform, replacing the 2G/3G units in earlier models. Those vehicles use the MyNISSAN app, which talks to a different backend."

Whenever you say A differs from B, name the part, the version, the date, the mechanism, or the supply-chain change that makes the difference real. If you do not have that detail, do not imply the difference exists.

Self-check before returning text¶

Run this pass on every piece of prose before you hand it back. The full banned lists are in references/ai-writing-detection.md; check against them directly.

Search for the emdash character. Remove every one (Rule 1).
Scan for banned verbs (delve, leverage, utilize, foster, bolster, underscore, unveil, streamline) and replace with plain equivalents.
Scan for banned adjectives and intensifiers (robust, comprehensive, pivotal, seamless, significantly, extremely, truly) and cut or replace.
Scan for banned transitions and openers (Furthermore, Moreover, That being said, In today's world, It's worth noting that).
Check every number: is it real and attributable? If not, cut it (Rule 2).
Check every sentence ends on a concrete detail, not an assertion of importance (Rule 5).
Check headings: does each name the content rather than tease it (Rule 16)?
Check for repeated points and repeated section shapes (Rules 6, 7).
Count hedging markers per paragraph. More than three is a red flag.
Read it aloud. If a phrase would sound unnatural to a colleague, rewrite it.

Full 24-rule reference¶

The complete non-negotiable list (including rules without worked examples above) is in CLAUDE.md in this folder. The rules not expanded above are:

Rule 2: No unsourced statistics — every number must be real and attributable.
Rule 3: No parenthetical clarifications in headings.
Rule 6: No repeated talking points — say it once.
Rule 8: Reference without narrating the reference — no "as discussed above".
Rule 9: No performative urgency without a reason — "Act now" needs a concrete consequence in the same sentence.
Rule 10: No scare quotes on normal words — quotation marks only for actual quotations from a named source.
Rule 12: Never start a sentence with "Whether you're."
Rule 14: No synthetic enthusiasm — no exclamation marks or cheerleading.
Rule 17: No fabricated case studies or scenarios.
Rule 18: No fabricated history or milestones.
Rule 20: No AI transition phrases (Furthermore, Moreover, Notwithstanding, That being said, At its core, In essence, It is worth noting that, In the landscape of, To put it simply).
Rule 21: No AI verbs (delve, leverage, utilize, facilitate, foster, bolster, underscore, unveil, navigate (metaphorical), streamline, endeavour, ascertain, elucidate).
Rule 22: No academic AI tells (shed light on, pave the way for, a myriad of, a plethora of, paramount, pertaining to, prior to, subsequent to, in light of, with respect to, in terms of, the fact that).
Rule 23: Quote sources accurately, and set off the long ones — every word inside quotation marks must match the source; name speaker and medium; long quotes (>~15 words) as indented blocks.
Rule 24: No research-process narration — silently omit what cannot be supported; do not write "could not be located" sections.