Audit content for AI-generated writing patterns and rewrite to sound authentic.
Scan Targets
- If paths provided: scan those paths
- Otherwise: scan markdown and source files project-wide (README*, docs/, src/, etc.)
Process
Phase 1: Scan and Flag
Vocabulary:
- Hype verbs: unleash, elevate, empower, foster, spearhead, unlock, delve, leverage
- Corporate adjectives: robust, comprehensive, seamless, cutting-edge, meticulous, pivotal
- Filler phrases: "in the rapidly evolving landscape of...", "it is important to note that..."
Structure:
- Emoji bullet lists with identical format per item (emoji + bold + explanation)
- Perfectly symmetrical lists (same length/structure for each item)
- Marketing tone for simple tools
Tone:
- Overly polite ("please feel free to...")
- Unwarranted enthusiasm
- Generic foo/bar examples lacking domain context
Phase 2: Report
Present findings: files scanned, flagged lines with pattern type, suggested rewrites.
Wait for approval before rewriting.
Phase 3: Rewrite
After approval, rewrite flagged content:
- Write like a dev explaining to another dev
- Use casual shorthand (repo, config, PR)
- Be direct about limitations
- Allow personality; skip polish
- No hype verbs, no symmetrical lists, no corporate fluff