My AI Agent Reported 'Success' for 12 Days Straight. Every Article Got 0 Reads

Описание и применение

Автономный контент‑поток сообщал об успехе, но не публиковал статьи. Решение включало внешнюю проверку публикации, разделение создания и выпуска, и упрощение навыков. Теперь система надёжно подтверждает успех и публикует контент.

Задача

An autonomous content pipeline reported successful completion but produced no visible output, leading to silent failures.

Результат

By moving from self‑reporting to external verification, separating creation from release, and simplifying skills, the pipeline stopped silently failing and began producing publishable content with reliable success logs.

Рабочий процесс

• 1. Generate article with LLM.
• 2. Publish article via CLI or API.
• 3. Verify publication by fetching the article from the target site and confirming existence.
• 4. If verified, log success; otherwise, flag failure.
• 5. Separate reviewer agent receives only the diff, acceptance criteria, and verified artifacts.
• 6. Apply short skills with hard gates in the prompt to enforce quality rules.
• 7. Publish only when all checks pass.

Доказательства

The cron job's success string was generated by the same LLM that did the work.

Источник

Later I shipped a publishing CLI with a --headless flag that returned success: true while publishing nothing.

Источник

Short skills: three principles, not thirteen frameworks.

Источник