petergyang/no-ai-slop

Purpose and delivery model

No AI Slop is a distributable agent skill for editing or detecting formulaic AI-writing patterns while retaining the author’s voice. It is not a conventional application: its primary runtime is a compatible AI agent (ChatGPT, Codex, Claude Code, or skill-aware CLI). The authored behavior lives in skills/no-ai-slop/SKILL.md; installable plugin metadata is in .codex-plugin/plugin.json.

User entry points

The primary invocation is the /no-ai-slop skill command, which accepts writing to revise or a detection-oriented prompt. The skill also supports a deliberate satire/generation mode for producing exaggerated “slop.” Agent discovery and integration are defined through:

• skills/no-ai-slop/SKILL.md

• agents/openai.yaml

• .codex-plugin/plugin.json

Core editing policy

SKILL.md is the system’s central specification and execution contract. It defines the catalog of unwanted patterns, voice-preservation constraints, editing workflow, and expected reporting of changes. The skill must distinguish between identifying patterns and asserting that text was AI-written; detection quotes patterns rather than inferring authorship.

• skills/no-ai-slop/SKILL.md

Pattern detection and revision stages

The skill’s main control flow is: receive text and user intent; classify the requested mode; inspect text for the defined slop patterns and general prose issues; either report matched language or revise it; then explain material edits. The pattern catalog includes binary contrasts, generic openers, faux-insight framing, reveal constructions, empty emphasis, puffery, vague attribution, synonym cycling, and manufactured profundity.

• skills/no-ai-slop/SKILL.md

• skills/no-ai-slop/eval.md

Quality boundary and evaluation

The evaluation document acts as a quality gate for the skill’s output. It checks that revisions remove target patterns without replacing them with equally generic phrasing, preserve concrete details and voice, and meet the required response behavior. This separates the editable instruction set from its acceptance criteria.

• skills/no-ai-slop/eval.md

• skills/no-ai-slop/SKILL.md

Packaging and distribution

The repository packages the skill as a Codex/ChatGPT plugin. Plugin identity, capabilities, and discovery metadata are declared in .codex-plugin/plugin.json. scripts/build_plugin.py builds and validates the package, creating a release boundary between authored skill files and a distributable artifact. GitHub Actions automates this plugin-oriented lifecycle.

• .codex-plugin/plugin.json

• scripts/build_plugin.py

• .github/workflows/plugin.yml

External ecosystem and governance

The project relies on external agent platforms and the npx skills installation ecosystem rather than hosting its own service or database. Its operational/legal boundary is documented by repository-level privacy and terms documents, while agents/openai.yaml provides provider-specific agent configuration.

• agents/openai.yaml

• PRIVACY.md

• TERMS.md