AI Writing Cleaner
Audits and rewrites prose to remove recognizable AI writing patterns.
The skill explicitly frames itself as a writing-quality aid rather than a verdict and warns about false positives and consequential decisions. It declares no external tools or APIs and shows no credential or covert-exfiltration path. Edit mode changes files in place; although it requires minimal edits and confirmation for large-file scope, it lacks general confirmation, backup, rollback, sensitive-data handling, and failed-edit recovery requirements, so points are deducted.
Modes, trigger phrases, output structures, and a second-pass audit are specified and broadly self-consistent. However, the skill provides no executable tests, abnormal-input handling, or clear failure messages. Under static calibration, execution-based confidence is unavailable, so reproducibility and diagnosability points are withheld.
Audience, rewrite/detect/edit modes, voice profiles, context profiles, and natural-language triggers are reasonably clear, and the non-verdict boundary is stated. Non-fit boundaries remain incomplete, Chinese-language support is unspecified, and several behaviors assume agent-specific tool capabilities, so points are deducted.
The frontmatter provides agentskills.io compatibility, version, MIT license, author, tags, modes, and parameter guidance, with generally clear information layering. The skill itself lacks installation guidance, change history, maintenance ownership, update path, FAQ, and systematic troubleshooting, and some capabilities depend on external agent tools, so points are deducted.
The skill directly specifies auditing, rewriting, change summaries, detection assessment, and post-edit rereading, so the core task is well defined. Static source review provides no representative end-to-end execution evidence within the selected skill, and the large rule set may cause over-editing and still require human review, so the score remains below the static maximum.
The rules include citations to research and source projects, and many claims are auditable from the text. However, the selected skill contains no test suite, CI execution evidence, or independently reproducible results, and citation coverage is incomplete, so only limited credit is justified.
- Edit mode modifies files in place without a universal backup, rollback, or pre-edit confirmation requirement; users should preserve a copy first.
- The rules may misclassify non-native English, technical genres, polished publications, or intentional personal style, and should not be the sole basis for academic-integrity, hiring, publication, or attribution decisions.
- Chinese-language coverage, false-positive behavior, and test results are not specified, so Chinese users should review outputs manually.
- The skill assumes an Edit tool and reread capability that may not exist or behave consistently across agents.
What does this skill do, and when should you use it?
This Agent Skills-compatible writing editor finds and removes common AI writing patterns, or “AI-isms.” It supports rewrite, detect-only, and in-place file-editing modes, with optional casual, professional, technical, warm, and blunt voice profiles. Rewrite mode includes a built-in corrective second pass and can iterate up to two passes when requested. It requires no external tools or APIs, although edit mode needs filesystem access.
Reads supplied text, or reads a named file in edit mode; identifies formatting, sentence-structure, vocabulary, promotional, chatbot, citation, URL, and other AI-writing patterns; returns an issue audit, rewritten text, change summary, and second-pass audit in rewrite mode; reports issues and contextual assessments in detect mode; or applies targeted file edits and verifies the result in edit mode. It accepts natural-language triggers and optional --mode, --voice, --context, --file, and --iterate settings.
- A content writer is preparing a blog post, social post, or marketing draft and wants to reduce machine-like phrasing without changing the core message.
- An editor needs to flag AI-writing patterns in published or third-party material without rewriting it.
- A developer wants an agent to clean a Markdown draft in place while preserving paragraphs that already sound natural.
- A team wants the same content rewritten in a casual, professional, technical, warm, or blunt voice.
- A reviewer needs to distinguish clear writing problems from patterns that may be intentional in context.
What are this skill's strengths and limitations?
- Covers formatting, sentence structure, vocabulary, document structure, chatbot artifacts, and other fingerprint-like signals.
- Offers detect-only and targeted in-place editing modes.
- Supports voice profiles and a corrective second pass after rewriting.
- Works with multiple SKILL.md-compatible agents and is MIT licensed.
- Its findings are signals, not proof; the skill explicitly warns about false positives, including for non-native English writers and technical genres.
- Edit mode depends on the host agent's ability to read, modify, and reread files.
- The source provides installation guidance but no independent testing evidence for every named platform.
- It edits writing patterns; it is not a fact-checker, authorship-attribution system, or definitive AI-generation detector.
How do you install this skill?
Place the repository's SKILL.md in the target agent's skills directory. For Claude Code, run: git clone https://github.com/conorbronsdon/avoid-ai-writing ~/.claude/skills/avoid-ai-writing . For OpenClaw, run: git clone https://github.com/conorbronsdon/avoid-ai-writing ~/.openclaw/skills/avoid-ai-writing . For Codex, run: mkdir -p .agents/skills/avoid-ai-writing && curl -o .agents/skills/avoid-ai-writing/SKILL.md https://raw.githubusercontent.com/conorbronsdon/avoid-ai-writing/main/SKILL.md . Other installation locations depend on the agent's support for SKILL.md or equivalent rule files.
How do you use this skill?
After installation, use prompts such as “Remove AI-isms from this post,” “Audit this draft for AI tells,” or “Edit draft.md in place.” Rewrite is the default. Use detect mode for requests such as “detect,” “flag only,” or “audit only.” Use edit mode when a named file should be changed directly. Optional settings include --voice casual|professional|technical|warm|blunt, --context linkedin|blog|technical-blog|investor-email|docs|casual, and --iterate with a maximum of two passes.
How does this skill compare with similar options?
Compared with a one-shot “make this sound human” prompt, it provides a structured audit, quoted issue identification, a change summary, detect-only mode, and a second-pass review. The source does not name other dedicated competitors.