De-AI Pipeline

Operator Context

Automated scan-fix-verify loop that removes AI writing patterns from documentation files. Uses scripts/scan-ai-patterns.py for deterministic detection against scripts/data/banned-patterns.json (323 patterns, 24 categories), then dispatches fix agents per file, then re-scans to verify fixes. Repeats until zero errors or max 3 iterations.

Hardcoded Behaviors (Always Apply)

• Script-First Detection: Use scan-ai-patterns.py for all scanning. Do not self-assess AI patterns.
• Preserve Meaning: Fixes must not change factual content. Only rephrase, never remove information.
• Skip Zones: Do not modify content inside code blocks, YAML frontmatter, inline code, or blockquotes.
• Max 3 Iterations: Stop after 3 scan-fix cycles even if errors remain. Report remaining errors.
• False Positive Awareness: When a pattern match is a technical term, skill name, or schema value, skip it and note the false positive.

Default Behaviors (ON unless disabled)

• Parallel File Fixes: When 2+ files have errors, dispatch fix agents in parallel.
• Commit After Clean: After reaching zero errors, stage and report (do not auto-commit).

Optional Behaviors (OFF unless enabled)

• Single File Mode: Pass a specific file path to scan and fix only that file.