Skip to content
Back to search

soundshuman

MarketplaceWorks with Claude CodeGitHub

The soundshuman skill and slop scanner, installable as a Claude Code plugin.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/aashaexo--soundshuman/badge.svg)](/listings/aashaexo--soundshuman)

HTML

<a href="/listings/aashaexo--soundshuman"><img src="/api/listings/aashaexo--soundshuman/badge.svg" alt="findsafeskills safety rating" /></a>

Install

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

As a skill, for any harness that reads markdown skills:

cp -r . /path/to/your/skills/soundshuman

As a Claude Code plugin:

/plugin marketplace add aashaexo/soundshuman
/plugin install soundshuman@soundshuman

The CLI needs nothing but Node:

node bin/sloplint.js --help

Or link it globally:

npm install -g .

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 17 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. {
  2. "$schema": "https://json.schemastore.org/claude-code-marketplace-manifest.json",
  3. "name": "soundshuman",
  4. "owner": {
  5. "name": "aasha"
  6. },
  7. "description": "The soundshuman skill and slop scanner, installable as a Claude Code plugin.",
  8. "plugins": [
  9. {
  10. "name": "soundshuman",
  11. "source": "./",
  12. "description": "Keep AI slop out of your prose and your repo: merged humanizer skill, repo auditing, and a slop scanner with CI gating.",
  13. "license": "MIT",
  14. "keywords": ["writing", "editing", "ai-detection", "humanize", "slop", "prose", "lint", "style"]
  15. }
  16. ]
  17. }
README.md · README · 120 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # soundshuman
  2. Keep AI slop out of your prose and your repo.
  3. This is a merger of the best parts of the current anti-slop tooling: one skill that rewrites text like a careful human editor, one rule pack that machines can enforce, and one zero-dependency CLI that audits whole repos and gates CI. The goal is not beating AI detectors. It is a workflow you can inspect: visible rules, explainable scores, reviewable diffs.
  4. ## What was taken from where
  5. | Source | What it contributed |
  6. |--------|---------------------|
  7. | [blader/humanizer](https://github.com/blader/humanizer) | The rewrite skill's backbone: the Wikipedia-derived pattern catalog, voice calibration from writing samples, the no-fabrication rule, false-positive guardrails, the draft, audit, final loop, and the plugin packaging. |
  8. | [hardikpandya/stop-slop](https://github.com/hardikpandya/stop-slop) | Structural anti-patterns the others miss (false agency, narrator-from-a-distance, binary contrasts, negative listing, lazy extremes, adverb pile), the pre-delivery checklist, and the lean skill-plus-references layout. |
  9. | [brandonwise/humanizer](https://github.com/brandonwise/humanizer) | The tooling ideas: tiered vocabulary, statistical tells (burstiness, type-token ratio, trigram repetition), explainable 0-100 scoring, repo-wide scanning with CI gates, and a test suite for the scanner. |
  10. | Aider | The editing workflow: rewrites land as git diffs on a clean tree, you review them like code, and undo is `git checkout`, not an apology. |
  11. | ast-grep / OpenRewrite | Rules as data. Detection lives in a versioned JSON rule pack, not in code, so a house rule is a reviewable one-line diff. |
  12. | Biome / Semgrep | Guardrails after the edit: a pre-commit hook and a CI job that catch what the session missed. |
  13. Everything here is MIT licensed; upstream copyrights are preserved in [LICENSE](LICENSE).
  14. ## The pieces
  15. [SKILL.md](SKILL.md) is the humanizer skill: 41 patterns across content, language, style, communication, filler, and rhetoric, plus statistical tells, voice calibration, and four invocation modes (pasted text, file, repo audit, embedded). It is plain markdown, so it runs in any agent harness that reads skills.
  16. [rules/slop-rules.json](rules/slop-rules.json) is the machine-readable rule pack: three vocabulary tiers, phrase fixes, and regex detectors with weights and notes. [bin/sloplint.js](bin/sloplint.js) is the scanner that reads it. Zero dependencies, Node 18+.
  17. The long-form material lives in [references/](references/): [vocabulary](references/vocabulary.md), [phrases](references/phrases.md), [structures](references/structures.md), a [style guide](references/style-guide.md), a [checklist](references/checklist.md), and [examples](references/examples.md). [docs/WORKFLOW.md](docs/WORKFLOW.md) explains how the pieces fit together, stage by stage.
  18. ## Install
  19. As a skill, for any harness that reads markdown skills:
  20. ```bash
  21. cp -r . /path/to/your/skills/soundshuman
  22. ```
  23. As a Claude Code plugin:
  24. ```
  25. /plugin marketplace add aashaexo/soundshuman
  26. /plugin install soundshuman@soundshuman
  27. ```
  28. The CLI needs nothing but Node:
  29. ```bash
  30. node bin/sloplint.js --help
  31. ```
  32. Or link it globally:
  33. ```bash
  34. npm install -g .
  35. ```
  36. ## Use
  37. Score any text (0 to 100, higher means more AI-flavored, under 25 is the target):
  38. ```bash
  39. echo "This serves as a testament to our seamless approach." | sloplint score
  40. ```
  41. See exactly why, with line numbers and the rule that fired:
  42. ```bash
  43. sloplint analyze draft.md
  44. ```
  45. Apply the safe mechanical fixes (filler phrases, em dashes, curly quotes):
  46. ```bash
  47. sloplint fix draft.md -w
  48. ```
  49. Audit a whole docs folder, worst first:
  50. ```bash
  51. sloplint scan docs/
  52. ```
  53. Gate CI:
  54. ```bash
  55. sloplint scan docs/ --fail-above 50
  56. ```
  57. Install the pre-commit hook:
  58. ```bash
  59. scripts/install-git-hook.sh 50
  60. ```
  61. For the full rewrite treatment, invoke the skill in your agent ("humanize this", "humanize docs/launch-post.md", "audit our docs for slop"). The skill drafts, audits its own draft, and delivers a final rewrite with no invented facts. Give it two paragraphs of your own writing first and it will match your voice instead of producing generic clean output.
  62. ## Always-on mode
  63. Prevention beats cleanup. To make an agent write like this by default, add the short version to your system prompt or CLAUDE.md:
  64. ```
  65. Write like a person, not a press release. Use "is" and "has", not "serves as"
  66. or "boasts". Never use: delve, tapestry, vibrant, crucial, robust, seamless,
  67. groundbreaking, pivotal, testament, leverage, empower, realm. No em dashes.
  68. No "It's not just X, it's Y". No "Great question!" or "I hope this helps".
  69. One qualifier per claim. Vary sentence length. End on a concrete fact, not
  70. "the future looks bright". Never invent facts, names, numbers, or citations.
  71. ```
  72. Then verify with `sloplint score` on the agent's output.
  73. ## Scoring, briefly
  74. Pattern density per 100 words on a log curve (70%), blended with statistical uniformity: burstiness, type-token ratio, and trigram repetition (30%). Bonuses for breadth and category diversity. Short samples get a confidence haircut instead of a confident wrong answer. Every point is traceable to a finding you can read, argue with, or delete from the rule pack.
  75. ## Development
  76. ```bash
  77. npm test
  78. ```
  79. The tests pin the scorer's behavior: AI samples score high, human samples score low, code blocks are ignored, fixes are safe. If you add a rule, add a test.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.