Add --source-mode to the bundled ChatGPT/Codex detector script

[conorbronsdon]

skills/ai-writing-detector/scripts/detect.js supports only --file and --context. scripts/sync-plugin-skill.sh copies patterns.js into that directory but never regenerates detect.js, so it has drifted from bin/avoid-ai-writing.js.

On a Markdown file with YAML frontmatter, the root CLI with --source-mode rendered-markdown reports [0, "Clean", []]. The bundled script reports [15, "Minimal AI signals", ["tier1", "tier1", "tier1"]], all three hits from the frontmatter. Passing the flag throws a raw Node stack trace (unknown argument: --source-mode) instead of the root CLI's exit-2 message.

Rendered-Markdown mode is active work (#123, #189, #190, #196) and the published plugin cannot reach it. The failure it prevents is the repo's worst false-positive class: unedited metadata scored as the author's prose.

Scope

Make detect.js a thin wrapper over the root CLI's parser, or add it to the files the sync script generates, so the two cannot drift again. Add --source-mode with matching validation and exit codes. Mirror the flag in skills/ai-writing-detector/SKILL.md with guidance on when to use it. Run the sync script and python3 scripts/validate-openai-plugin.py . --json.

Acceptance criteria

• The bundled script gives identical results to the root CLI for the same flags.
• An unknown flag exits 2 without a stack trace.
• The sub-skill's SKILL.md documents --source-mode.
• Either the sync script generates detect.js (git diff empty after running it) or a test pins parser parity.
• npm test and the OpenAI plugin validator pass.

Add an ## [Unreleased] changelog entry: this changes the published plugin.

[conorbronsdon]

Opening this to first-time contributors, with the implementation choice made so nobody has to guess.

Reproduced on current main (fabd62d): bin/avoid-ai-writing.js --source-mode rendered-markdown <file> returns Clean on a Markdown file with YAML frontmatter, while skills/ai-writing-detector/scripts/detect.js --file <same file> flags all three frontmatter lines, and passing it --source-mode throws an uncaught unknown argument error with a stack trace.

Decision: keep detect.js's existing --file and --context interface as it is. Do not try to unify the two CLIs (the root CLI takes a positional path, detect.js takes --file; that difference stays). Just port the one flag.

What to do:

1. In skills/ai-writing-detector/scripts/detect.js, add --source-mode <plain|rendered-markdown>, validate the value, and pass it through as sourceMode to analyzeText. The masking already lives in detector/patterns.js, so no detector change is needed.
2. Replace the throw new Error(...) on a bad argument with a usage message and process.exit(2), matching the root CLI.
3. Add a parity test modeled on bin/avoid-ai-writing.test.js (it spawns the CLI and compares JSON against detector/patterns.js directly): the same content through both CLIs with --source-mode rendered-markdown must produce the same issues. Register the new test file in scripts/run-tests.js.
4. Document the flag in skills/ai-writing-detector/SKILL.md next to --context.
5. Verify with npm test and python3 scripts/validate-openai-plugin.py . --json, then run bash scripts/sync-plugin-skill.sh && bash scripts/sync-cursor-rules.sh and confirm the working tree is clean apart from your edits.
6. Add an ## [Unreleased] changelog line: the bundled detector script accepts --source-mode.

---

Generated by Claude Code