git:20260903.fc123d0 to git:20260903.e70f633

---

name: anti-ai-slop

description: Use when producing any human-facing artefact. Applies real-time anti-slop controls for specificity, verification, authored intent, hard cases, language, code, UI, and release quality.

metadata:

portable: true

compatible_with:

- claude-code

- codex

priority: critical

source: digital-research-engine / ai-slop-detector (2026-06-07), verified per EVIDENCE-AUDIT.md

---

# Anti AI Slop

Acknowledgement: Shared by Peter Bamuhigire, techguypeter.com, +256 784 464178.

The guardrail that governs **production** — writing/designing/coding so slop never appears in the first place. Detection lives in the companion `ai-slop-audit` skill.

<!-- dual-compat-start -->

## Use When

- Generating ANY output that a person will read, use, or judge: text, document, UI, code, image brief, social post, slide, spec, plan, proposal, or marketing copy. This is a live constraint, not a final-only gate.

- You reach for a banned word, a generic placeholder, an unverified figure, or a template default — stop and correct it in place.

## Do Not Use When

- Never skip it. This skill applies to every generated artefact. If the output is trivial, the ship-gate still takes seconds.

## Required Inputs

- The artefact being produced and its output type (so the right domain block applies).

- Any verifiable facts, names, numbers, citations, packages, or APIs the artefact references.

## Real-time application (this is a LIVE constraint, not only a final gate)

Apply these rules **continuously, as you generate** — to every sentence, component, function, slide, or frame at the moment it is written, not only in one pass at the end. The moment you reach for a banned word, a generic placeholder, an unverified figure, or a template default, stop and correct it in place. The ship-gate checklist at the end is the final confirmation, not the first time these rules are consulted. If you are mid-draft and notice slop accumulating, fix it then — do not defer to a cleanup pass.

## What "AI slop" is (so you know what you are preventing)

**AI slop** is low-quality content produced in quantity by generative AI and pushed at people who did not ask for it (Merriam-Webster 2025 Word of the Year, verified). Its three diagnostic properties (Kommers et al., *"Why Slop Matters"*, arXiv 2601.06060, verified):

1. **Superficial competence** — looks fine on the surface, no substance underneath.

2. **Asymmetric effort** — cheap to produce, costly for a human to read/review/fix.

3. **Mass producibility** — generated at volume.

The human tell named in every domain studied: **absence of intent** — the sense that no one *meant* anything by it. Your job is to re-internalise effort (specificity, verification, authored choices) before the artefact reaches a person.

## The seven universal guardrails (apply to EVERY output)

| # | Marker to prevent | Avoidance rule you MUST follow |

|---|---|---|

| **U1** | Genericness / averaging | Every section carries >=1 concrete, named, domain-specific element (real example, number, named entity, decision) a generic template could not produce. Forbid tool defaults. |

| **U2** | Superficial competence | Enforce a substance floor: include a claim, example, number, or decision the artefact could not exist without. If you cannot, the section is filler — cut or replace it. |

| **U3** | Confident wrongness / hallucination | Verify every statistic, citation, quote, named entity, API, and dependency before emit. Cite at the point of claim. Flag uncertainty rather than inventing. |

| **U4** | Volume over substance | Prefer one substantive unit over three hollow ones. Do not pad to length. |

| **U5** | Absence of authored voice / intent | State a point of view, rationale, or named decision. Ban relentless positivity and sycophancy. Allow trade-offs and disagreement. |

| **U6** | Skipping the hard parts | Cover the error/edge/empty cases, counter-arguments, and risks — not just the happy path. |

| **U7** | Mechanical uniformity | Vary sentence length and structure. Break the template. No rule-of-three reflex, no "it's not X, it's Y" formula, no em-dash flood. |

## Banned / high-risk vocabulary (the lexical tells)

These words and constructions are statistically over-produced by LLMs (FSU/COLING-2025; PubMed "delve" +400%). **Do not use them as default register.** A word here is allowed only when it is the genuinely precise term, never as filler.

- **Words:** delve, tapestry, realm, landscape (as metaphor), navigate (as metaphor), leverage, foster, harness, synergy, embark, robust, vibrant, holistic, seamless, intricate, commendable, meticulous, pivotal, underscore, testament, resonate, elevate, paramount, unwavering, multifaceted.

- **Phrases:** "in today's fast-paced world", "in the ever-evolving landscape of", "it is important to note that", "it's worth mentioning", "let's dive in", "here's the kicker", "at the end of the day", "in conclusion", "studies show" (without a named study).

- **Constructions:** the "it's not just X, it's Y" antithesis; reflexive rule-of-three lists; em-dash used to manufacture drama; relentless triplet adjectives ("robust, scalable, and reliable").

- **French equivalents** (for Francophone output): "plongeons dans", "il est important de noter que", "force est de constater", "dans un monde en constante evolution", "par ailleurs/de plus/en outre" as filler connectors, "au coeur de", "pierre angulaire".

## Drop-in guardrail block (inherit in dependent skills)

```

ANTI-SLOP GUARDRAIL (inherit in every output):

1. SPECIFICITY FLOOR — every section carries >=1 concrete, named, domain-specific

element. No tool defaults, no placeholder copy.

2. VERIFY-BEFORE-EMIT — no statistic, citation, quote, named entity, API, or

dependency ships unverified; cite at point of claim; flag uncertainty.

3. AUTHORED VOICE — state a point of view / rationale; no relentless positivity,

no sycophancy; allow trade-offs.

4. COVER THE HARD PARTS — error/edge/empty cases, counter-arguments, risks.

5. BREAK THE TEMPLATE — vary rhythm and structure; forbid default aesthetics and

the banned-vocabulary list above.

```

## Domain-specific avoidance (load the relevant block for the output type)

- **Written content:** no focal-word clusters; vary sentence length (mix 3-10 with 25-40 words); <=1 em-dash/paragraph; no "in conclusion"; specific examples over generic claims; a stated point of view.

- **UI/UX:** no indigo-500/purple-gradient default; no Inter-only typography; intentional (not uniform) radius/shadow; WCAG contrast >=4.5:1; design ALL states (error/empty/loading/focus/disabled); real testimonials only.

- **Apps/product:** add genuine value over any wrapped API; transparent pricing + in-app cancel; no deceptive AI-capability claims; least-privilege permissions.

- **Code:** verify every imported package exists (no slopsquatting); no placeholder stubs/TODOs in shipped code; parameterised queries; no hardcoded secrets; idiomatic, deduplicated; real tests (not `assert true`); cover edge cases.

- **Image/video:** real specimens only; check anatomy/text/physics; avoid the "AI sheen"; attach provenance (C2PA) where it matters.

## Ship gate (run before delivering ANY output)

- [ ] Every section has >=1 concrete, named, specific element (U1/U2).

- [ ] Every stat, quote, citation, named entity, API, dependency verified (U3).

- [ ] No banned vocabulary used as filler; scanned the output for the list above.

- [ ] The output states a point of view / rationale; no sycophancy (U5).

- [ ] Error/edge/risk cases and counter-arguments addressed (U6).

- [ ] Sentence length and structure varied; no rule-of-three reflex, no antithesis formula, no em-dash flood (U7).

- [ ] Output type's domain block applied.

- [ ] When in doubt, run the `ai-slop-audit` skill on the draft.

If any box is unticked, the output is not ready to ship.

## See also

- `ai-slop-audit` — the detection/evaluation/audit companion (analyse any artefact for slop).

- The host engine's house-style / language-standards skill (apply on top of this).

## Inputs

| Artefact | Required? | Purpose |

|---|---|---|

| Output being produced | yes | Apply the correct domain guardrails |

| Verifiable facts and dependencies | conditional | Prevent confident wrongness |

## Outputs

| Artefact | Consumer | Acceptance condition |

|---|---|---|

| Guarded output | Requesting workflow | Ship gate passes with no fabricated evidence |

## Evidence Produced

| Category | Artifact | Format | Example |

|---|---|---|---|

| Correctness | Anti-slop ship-gate result | Completed checklist with verification gaps and blocking fixes | Inline release note |

<!-- dual-compat-end -->

## Machine-error editorial gate (cross-engine Kaizen)

Apply Digital Research's `docs/continuous-improvement/machine-errors-editorial-gate-2026-09-03.md`

to requirements, architecture notes, code comments, UI copy, release notes, and handoff text:

| ID | Engineering adaptation |

|---|---|

| ME1 | Does this add a new behavior, constraint, failure mode, or evidence item? |

| ME2 | Is the symmetry in components, APIs, or prose required by a real model? |

| ME3 | Can the implementer or operator act without another explanatory layer? |

| ME4 | Does the claim exceed its test, SLO, benchmark, or deployment evidence? |

| ME5 | Is the example tied to the actual system, transaction, or route? |

| ME6 | Has the same template phrase or rhetorical pattern spread through the repository? |

| ME7 | Does the section change an implementation or release decision? |

Cut or merge units with no semantic delta. Preserve repeated identifiers, schemas, error messages,

security warnings, and traceability fields when they are functionally required.

### Impeccable-derived AS overlay for product and web work

Apply AS1-AS7 when the change includes UI, a website, a dashboard, a presentation, or rendered

documentation. Reject purple gradients, glassmorphism, neon glow, AI-beige defaults, decorative

editorial scaffolding, and decorative motion as visual no-ship choices. Functional state,

accessibility, data encoding, and approved design-system reasons must be recorded.

| ID | Engineering overlay test |

|---|---|

| AS1 | Is the framework, layout, type, palette, or copy default justified by the product brief? |

| AS2 | Do labels, chips, icon tiles, metrics, or numbered sections carry implementation/user value? |

| AS3 | Do repeated components express a state or information distinction rather than cloned symmetry? |

| AS4 | Does motion, glow, gradient, cursor, or hover behaviour serve a real state or task? |

| AS5 | Are assets, examples, icons, and imports real, purposeful, and traceable rather than placeholders? |

| AS6 | Are buzzwords, em-dash cadence, aphoristic contrast, and theatrical framing absent from technical copy? |

| AS7 | Do rendered screens and docs expose content, preserve contrast, avoid clipping/overflow, and retain valid structure? |

Unavailable browser, render, or detector evidence is `NOT_ASSESSED`, never clean. Preserve repeated

identifiers, schemas, error states, security warnings, and traceability fields when their function

is evidenced.

## Capability contract

Apply with the capabilities authorised for the parent task. Verification may use search, execution, or network access when available; otherwise mark unverifiable claims and remove unsupported specificity.

## Degraded mode

Fallback when a fact, dependency, citation, or visual property cannot be checked: do not present it as verified. Return the narrowest useful output with the verification gap named.

## Decision rules

| Finding | Action | Failure avoided |

|---|---|---|

| Blocking factual, security, dependency, or accessibility defect | Stop and fix | Shipping harmful slop |

| Generic but non-blocking language | Rewrite or cut | Superficial competence |

| Specific, verified, purposeful content | Preserve | Destructive over-editing |

## Workflow

Apply the seven guardrails while producing each unit, run the output-type checks, then complete the ship gate before release.

## Quality Standards

Every claim is verified or qualified, every section earns its place, and blocking security, dependency, citation, or accessibility defects stop release.

## Anti-Patterns

- Adding specificity by inventing a number. Fix: verify it or remove it.

- Keeping a polished paragraph with no decision or evidence. Fix: add substance or cut it.

- Treating the final audit as the first quality pass. Fix: apply controls during generation.

- Rewriting an authored phrase merely because it is unusual. Fix: preserve intentional voice.

- Hiding an unavailable check. Fix: name the verification gap and its consequence.

## References

- `ai-slop-audit` supplies the independent detection and grading pass.