What problem does it solve?
AI-generated technical prose often contains clichés, filler phrases, passive voice, and vague abstractions that erode reader trust. This Skill provides a curated ruleset of 21 writing rules, distilled from Strunk & White, Orwell, Pinker, and Gopen & Swan, so agents can write and revise formal technical prose that reads clearly and credibly.
Core Features & Use Cases
- 21-Rule Ruleset: Twelve canonical rules (curse of knowledge, passive voice, needless words, dying metaphors, claim calibration, and more) plus nine field-observed rules targeting LLM-specific tells like bullet overuse, em-dash overuse, and summary closers.
- Full Rule Bodies with Examples: The vendored references/RULES.md snapshot provides directives, BAD/GOOD example pairs, severity ratings, and rationale for each rule.
- Optional Deterministic Audit: Integrates with the upstream agent-style CLI for mechanical rule auditing of files.
- Use Case: Ask the agent to revise a design doc, README, or commit message per agent-style rules, and it will cut filler phrases, replace clichés with concrete numbers, and calibrate claims to evidence.
Quick Start
Ask the agent to revise the attached design document per agent-style rules and iterate until it is agent-style clean.