Teaching Agents to Recognize AI Slop: The Failure Library

Where this fits

You're reviewing a draft your agent just produced for a client blog post. The prose is grammatically perfect but somehow lifeless—every sentence ends with a neat conclusion, transitions feel rehearsed, and the metaphors are suspiciously symmetrical. You know it's AI slop but can't articulate why. The failure-library skill gives you (and your agent) a vocabulary for diagnosing these patterns, turning vague unease into a precise, teachable breakdown.

Why agents benefit

• Diagnoses with evidence: Instead of guessing why a passage feels machine-written, the agent can match it against documented failure patterns and cite the exact mechanism.
• Teaches through contrast: When refining its own output, the agent can reference real examples of what to avoid, not just abstract rules.
• Explains to humans: The agent can generate clear, annotated explanations of why a sentence fails, helping editors and writers learn rather than just fix.
• Prevents over-correction: The skill's core rules (anti-ai-rules.md) explicitly warn against overcorrecting into stilted prose, so the agent learns balance.
• Works as a reference or a live tool: It can be read standalone for teaching or loaded during generation to guide real-time edits.

Practical scenarios

Editing a newsletter draft

Your agent drafts a weekly tech newsletter. You notice every paragraph starts with a topic sentence and ends with a takeaway. The agent loads the failure library, identifies the "forced structure" pattern, and rewrites with varied rhythm and deliberate asymmetry.

Teaching a junior writer

A new writer keeps producing generic transitions like "Moreover" and "In conclusion." You ask the agent to pull up the library's "transition filler" entry and generate a mini-lesson with three before/after examples from real AI text.

Auditing a batch of AI-generated content

You have 50 product descriptions from an external vendor. The agent scans them, flags passages that match known failure patterns (e.g., "hedging with synonyms"), and produces a diagnostic report with severity ratings and suggested fixes.

Add it to your agent workflow

Install the skill directly:

npx skills add Deupaxx/EveryDay-Writer --skill failure-library

Then, in your agent's system prompt, add a line like:

When editing or generating text, consult the failure-library skill to diagnose and avoid AI writing patterns. Read core/anti-ai-rules.md first.

For a quick diagnostic, ask: "Check this paragraph against the failure library. Identify any patterns and explain the mechanism."

Compare before adopting

Before committing, compare the failure library against other writing-quality skills on these axes:

• Quality signals: Does it use real, annotated examples or just generic advice? The library's provenance matters.
• Maintenance freshness: Is the skill actively updated? Check the repo's commit history and issue activity.
• Alternatives: Other skills focus on style guides or grammar checks—this one is specifically about AI-slop patterns. Make sure that's your actual need.
• Workflow fit: Does it integrate with your agent's existing rules? The skill expects a specific reading order, so verify it doesn't conflict with your current setup.

Why it is worth tracking

With 28 stars and an MIT license, this skill is young but well-scoped. Its value lies in the quality of its dissections—if the library grows with fresh, real-world examples, it becomes an indispensable reference for any agent that writes. Keep an eye on how often it's updated and whether the author adds new failure categories. Evaluate it when you're tired of generic writing tips and want something that actually teaches your agent to see the difference between human and machine prose. For more context, see the skill listing at OpenAgentSkill.