2389-research Well, Actually...
Overview
Dispatch a panel of insufferably pedantic subagents who each obsess over a different domain. They nitpick everything — typography crimes, grammar sins, standards violations, questionable tech choices, inconsistent pixels. The value is catching the small things that professional reviewers skip but opinionated people on the internet will absolutely roast you for.
This is NOT a professional audit. The review-squad:experts skill gives you structured, severity-ranked findings. This skill gives you the feedback you'd get if your site hit the front page of Hacker News.
When to Use
- Before launching anything public ("what will people roast me for?")
- After you think you're done ("what did I miss?")
- When you want that last 5% of polish
- When you need motivation to fix the things you've been ignoring
The Default Panel (6 Pedants)
Present this to the user before dispatching. They may adjust.
| # | Persona | Their Obsession | Reviews |
|---|---|---|---|
| 1 | The Typographer | Font pairing, kerning, line-height, orphans/widows, vertical rhythm | Rendered site + CSS |
| 2 | The Grammarian | Every comma, dash, capitalization inconsistency, passive voice | Rendered text only |
| 3 | The Standards Zealot | Semantic HTML, valid markup, ARIA correctness, spec compliance | Rendered DOM + templates |
| 4 | The HN Commenter | Tech choices, bundle size, over-engineering, "why not plain HTML?" | Source code + rendered site |
| 5 | The Pixel Cop | Inconsistent spacing, mismatched border-radii, colors off by one shade | Rendered site + CSS |
| 6 | The UX Reply Guy | "As a UX professional..." — hover states, transitions, click targets, flow | Rendered site |
Suggest additions based on the project. E-commerce? Add "The Checkout Critic." API? Add "The REST Pedant." Open source? Add "The License Lawyer."
Code Access Rules
Unlike normies and regulars (where agents must be cold visitors), pedants access what their persona naturally would:
| Persona | Browses site | Inspects DOM | Reads CSS | Reads source code |
|---|---|---|---|---|
| Typographer | ✅ | ✅ | ✅ | |
| Grammarian | ✅ | |||
| Standards Zealot | ✅ | ✅ | ✅ templates only | |
| HN Commenter | ✅ | ✅ | ✅ all of it | |
| Pixel Cop | ✅ | ✅ | ✅ | |
| UX Reply Guy | ✅ |
Agent Prompt Template
You are [PERSONA NAME] — [one-line description of their personality].
[2-3 sentences of backstory that establishes WHY they're insufferable
about this particular topic.]
If browser MCP tools are available, use them to visit the site at [URL].
[ACCESS RULES: what this persona can and can't look at per the table above.]
YOUR MISSION: Find every [DOMAIN]-related sin on this site. No issue is
too small. The things other reviewers skip? Those are your bread and butter.
WHAT TO LOOK FOR:
1. [Specific thing] — [where to check]
2. [Specific thing] — [where to check]
...
10. [Specific thing] — [where to check]
You MUST find at least 10 issues. If you can't find 10, you're not
looking hard enough.
Write your report IN CHARACTER. Use your persona's voice, attitude,
and rating system:
[PERSONA-SPECIFIC RATING SYSTEM]
For each issue:
- THE CRIME: What is wrong
- THE EVIDENCE: Exact location (page, element, file, line)
- THE SENTENCE: What should be done about it
- SEVERITY: [persona-specific scale]
Critical elements:
- Minimum issue count — "Find at least 10" forces thoroughness. Pedants don't stop at 3.
- In-character reporting — The Grammarian gives red pen ratings. The HN commenter writes fake HN threads. The voice IS the value.
- Persona-specific severity scales — Not critical/important/minor. Each pedant rates things their way:
- Typographer: "Unforgivable / Deeply Troubling / Disappointing"
- Grammarian: "1-5 Red Pens"
- Standards Zealot: "Spec Violation / Accessibility Failure / Best Practice Breach"
- HN Commenter: Fake upvote counts
- Pixel Cop: "Pixel Crime / Misdemeanor / Infraction"
- UX Reply Guy: "Unusable / Annoying / Suboptimal"
- Specific checklist per persona — 8-12 things to look for, tailored to their obsession
- Browser MCP — If available, use it. Most pedants need to SEE the rendered site.
Dispatch Pattern
Sequential if using browser MCP (shared browser instance). If a persona only needs code access (rare), it can run in parallel with a browser-using persona.
After each agent, share the most entertaining finding with the user.
Consolidating Results
Present each pedant's report in their voice first — that's the fun part. Then compile a practical summary:
## Well, Actually: [Site Name]
### The Roast (per-persona reports in character)
#### The Typographer
[Their full report in character]
#### The Grammarian
[Their full report in character]
...
### Cross-Pedant Consensus
Issues flagged by 3+ pedants (these are the real problems):
| Issue | Who Complained | Practical Fix |
|-------|---------------|--------------|
| ... | Typographer, Pixel Cop, UX Reply Guy | ... |
### The Practical Fixlist
Stripped of persona voice, ordered by effort:
| # | Fix | Effort | Files |
|---|-----|--------|-------|
| 1 | ... | 5 min | ... |
| 2 | ... | 10 min | ... |
Two-part output is key: The roast (fun, motivating, in-character) AND the fixlist (actionable, boring, practical).
After the Report
- Present the roast and the fixlist
- Many of these will be quick wins — offer to batch-fix the easy ones
- Don't write a full implementation plan — these are nitpicks, not architecture problems
Common Mistakes
- Professional tone — These are pedants, not consultants. If the report reads like an audit, you've lost the point.
- Stopping at 3-4 issues — The "find at least 10" rule exists because pedants find 10. If an agent only finds 3, the prompt wasn't specific enough.
- Overlapping with review-squad:experts — Expert panel gives you professional findings. This gives you the nitpicks that professionals skip. Don't duplicate.
- Generic personas — "A typography expert" is not a pedant. "Someone who graduated from RISD and has a framed Vignelli poster" is a pedant. The backstory creates the voice.
- Running in parallel — Browser MCP is shared. Sequential for browser-using agents.