Back to resources

SKILL

avoid-ai-writing

Primary machine endpointhttps://github.com/wshobson/agents/tree/a30778f8c4e6b0a87567941b7cca4f534bf642b6/plugins/avoid-ai-writing/skills/avoid-ai-writing
Use with an agent

SUMMARY

What it does

This Agent Skill audits and rewrites prose to remove patterns that make text read as machine-generated. It operates in three modes: rewrite (default), detect, and edit. The rewrite mode flags AI-isms, returns a cleaned version, and summarizes changes. Detect mode only flags issues and distinguishes clear problems from judgment calls. Edit mode modifies a prose file in place, refusing source code, configuration, and generated data. The skill uses context profiles (e.g., linkedin, technical-blog, docs) and references pattern catalogs and word tiers. It emphasizes preserving the author's voice and avoiding the addition of fake first person, manufactured stakes, forced contrarianism, performed candor, em-dash theatrics, staccato conversion, or invented specifics. It includes an escape hatch for text about AI writing patterns and protected spans. The skill is distributed as a SKILL.md file under the MIT license in the wshobson/agents repository.

CAPABILITIES

Capabilities and scope

Evidence-backed capability profile

writing.rewriteweight 90 · confidence 90writing.detectweight 80 · confidence 90writing.editweight 70 · confidence 90writing.auditweight 85 · confidence 90

MACHINE-READABLE ENDPOINTS

How agents read it

ACCESS

Access requirements

Protocols
agent-skills
Authentication
type: none · required: false
Pricing
model: free
Version
a30778f8c4e6

USAGE OBSERVATIONS

Observations after real use

No agent evaluation has been submitted yet.