anti-ai-slop

name	anti-ai-slop
description	Detect and improve AI slop, generic-output, template-like, and AI-smell patterns in writing, UI, slides, visuals, code, reports, and other artifacts without making authorship claims.
tags	["codex-skill","spec-kit","guardrails","ai-slop","output-quality","prompt-engineering","code-review","documentation-review","validation","agent-workflows"]

Anti-AI Slop Skill

Use this skill to evaluate whether an output looks purpose-built or merely generic, over-smoothed, template-like, or AI-scented.

This skill is not an AI detector. It does not decide whether something was generated by AI. It identifies visible output patterns that can reduce credibility, specificity, usability, or brand fit.

Core principle

A weak output usually fails because it optimizes for the average shape of a good answer rather than the actual job to be done.

Review every output against three questions:

1. Purpose: What job is this output supposed to perform?
2. Specificity: What details make it unmistakably fit this user, audience, product, context, or constraint?
3. Design quality: Does the structure, wording, interface, visual system, or code support the purpose without decorative noise?

When to use this skill

Use this skill when the user asks to:

• make writing, UI, slides, visuals, or code feel less generic or less AI-generated;
• review a draft, landing page, deck, report, product concept, design system, or generated artifact;
• create output-quality criteria, QA checklists, review rubrics, or acceptance criteria;
• add output design review considerations to a deliverable;
• classify common AI-smell patterns by work purpose;
• improve credibility, originality, specificity, brand voice, or human editorial judgment.

Do not use this skill for

• forensic authorship detection;
• accusing a person or organization of using AI;
• replacing factual verification, legal review, security review, accessibility review, or domain expert review;
• removing useful structure only because it resembles an AI pattern.

Non-negotiable review rules

1. Never claim AI origin from style alone. Say “generic pattern,” “AI-smell risk,” or “template-like signal,” not “this was written by AI.”
2. Preserve functional clarity. Bullets, tables, simple words, and clean UI are not defects when they serve the task.
3. Prioritize purpose over taste. A blinking status dot, Inter-like font, or three-card grid can be correct when justified by the product and interaction model.
4. Base findings on visible evidence. Every critique should point to a phrase, layout decision, component, claim, or code pattern.
5. Pair every issue with a fix. Do not only diagnose. Provide concrete revision instructions or replacement examples.
6. Respect user constraints. Keep requested tone, format, medium, language, compliance limits, and artifact requirements.

Skill routing

For repository-level guardrail work, start with:

• .specify/memory/constitution.md
• .specify/memory/glossary.md
• .specify/memory/architecture-principles.md
• .specify/memory/product-principles.md
• codex/prompts/00-apply-pack-to-existing-repo.md

Then run or recommend:

python tools/validators/check_all.py --root . --format markdown

For single-artifact review work, start with the original review flow:

Start with:

• protocols/review_workflow.md
• taxonomies/task_purpose_matrix.md
• protocols/output_design_review_gate.md

Then load the relevant dimension file:

Output type	Use
General explanatory writing, summaries, documentation	dimensions/writing_information.md
Academic, research, policy, analytical, or executive reporting	dimensions/academic_reporting.md
Marketing copy, brand language, landing pages, sales material	dimensions/marketing_brand.md
Essays, social posts, founder notes, thought-leadership writing	dimensions/emotional_social.md
Websites, apps, SaaS UI, design systems, components	dimensions/web_ui_design.md
Slide decks, pitch decks, lecture decks	dimensions/presentation_decks.md
Generated images, thumbnails, visual concepts, hero imagery	dimensions/images_visuals.md
Code, PRs, generated frontend, scripts, technical snippets	dimensions/code_developer_outputs.md
Charts, dashboards, analytics screenshots, BI mockups	dimensions/data_charts_dashboards.md
Localization, translated copy, multilingual UI, region-specific outputs	dimensions/localization_register.md
Roadmaps, implementation plans, launch plans, operational specs	dimensions/operational_plans.md

Use these cross-cutting checklists when relevant:

• checklists/global_ai_smell_checklist.md
• checklists/remediation_patterns.md
• checklists/output_design_addendum.md

Use field-reported pattern research when the user asks about AI-smell, AI slop, generic-output tells, Reddit/GitHub-reported patterns, detector-like style cleanup, PR/code slop, or emerging public perception patterns:

• research/field_reported_ai_smell_patterns.md

Use templates for final deliverables:

• templates/review_report_template.md
• templates/design_review_addendum.md
• templates/rewrite_brief_template.md

Use validator tools for repository coherence:

• tools/validators/check_all.py
• tools/validators/check_config_integrity.py
• tools/validators/check_task_traceability.py
• tools/validators/check_spec_coverage.py
• tools/validators/check_evidence_links.py
• tools/validators/check_glossary_terms.py
• tools/validators/check_architecture_boundaries.py
• tools/validators/check_dependency_justification.py
• tools/validators/check_manifest_integrity.py
• tools/validators/check_resource_catalog_freshness.py
• tools/validators/check_template_tokens.py
• tools/validators/check_workflow_integrity.py
• tools/validators/check_vague_language.py
• tools/validators/check_placeholders.py

Repository guardrail helpers:

• tools/repo_inventory.py
• tools/bootstrap_feature.py
• tools/apply_guardrails.py
• tools/resource_catalog_freshness.py
• dependency-baseline.json

check_all.py supports --profile pack-self, --profile target-repo, --profile feature --feature FEATURE_DIR, and --profile ci-strict. Optional shared validator config may live at config/guardrails.json or config/guardrails.yaml; local config/guardrails.local.json overlays are rejected until merge semantics are implemented. Use --manifest-out with tools/apply_guardrails.py when applying the pack to another repository. Dry-run mode must not write reports or manifests inside the target repository; use --json-out outside the target when a dry-run report is needed. Apply plans use the maintained manifest.txt surface while excluding test-only files and generated cache artifacts such as tests/, __pycache__, *.pyc, and .pytest_cache.

Standard review workflow

1. Define the output contract. Identify audience, purpose, format, constraints, and success condition.
2. Classify the work purpose. Use taxonomies/task_purpose_matrix.md.
3. Run the output design gate. Use protocols/output_design_review_gate.md.
4. Run modality-specific checks. Use the relevant file under dimensions/.
5. Score only what is useful. Use severity and confidence from protocols/finding_format.md.
6. Recommend fixes. Use checklists/remediation_patterns.md.
7. Produce the final output or review report. Use the appropriate template.

Default finding format

### P1 — Generic value proposition
- Evidence: The hero says “Supercharge your workflow” without naming the actual user task.
- Why it matters: The sentence could apply to any SaaS product and does not create positioning.
- Fix: Replace the abstract benefit with a user-specific task, measurable result, or concrete workflow.
- Example revision: “Turn support-call transcripts into QA-ready coaching notes in under 3 minutes.”

Package layout

anti-ai-slop/
├── AGENTS.md
├── SKILL.md
├── README.md
├── VERSION
├── CHANGELOG.md
├── dependency-baseline.json
├── .specify/
├── .agents/
├── codex/
├── docs/
├── specs/
├── tools/
├── tests/
├── manifest.txt
├── protocols/
│   ├── review_workflow.md
│   ├── finding_format.md
│   └── output_design_review_gate.md
├── taxonomies/
│   └── task_purpose_matrix.md
├── checklists/
│   ├── global_ai_smell_checklist.md
│   ├── remediation_patterns.md
│   └── output_design_addendum.md
├── research/
│   └── field_reported_ai_smell_patterns.md
├── dimensions/
│   ├── writing_information.md
│   ├── academic_reporting.md
│   ├── marketing_brand.md
│   ├── emotional_social.md
│   ├── web_ui_design.md
│   ├── presentation_decks.md
│   ├── images_visuals.md
│   ├── code_developer_outputs.md
│   ├── data_charts_dashboards.md
│   ├── localization_register.md
│   └── operational_plans.md
├── templates/
│   ├── review_report_template.md
│   ├── design_review_addendum.md
│   └── rewrite_brief_template.md
└── examples/
    └── before_after_patterns.md

Maintenance notes

For pack-level validator, workflow, migration-tool, or guardrail behavior changes, update VERSION, CHANGELOG.md, and manifest.txt with the same change.