Board
Skill mickzijdel/dev-hooks/plugins/thinking-tools/skills/board
Hooks and skills for Claude to write better code and verify its work, and an easy start to using Claude Code
npx -y skills add mickzijdel/dev-hooks --skill boardAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use when you want hard, independent, adversarial critique of an idea, plan, design, draft, proposal, pitch, or landing page — especially when you suspect the default response is too agreeable or sycophantic and you want real pushback before committing. Convene it on your own initiative before returning a consequential plan or draft, not only when asked. Triggers on "poke holes in this", "what am I missing", "be brutal", "convince me this is wrong", or any decision that benefits from several expert angles at once.
SKILL.md
3.7 KB, as published. Nobody here has run it
Board of Advisors
Convene a panel of independent advisors who each attack the work from a different angle, then a chairman synthesizes. The point is to defeat agreement bias: real, separate critics surface problems a single agreeable pass glosses over.
Engine: spawn the advisors as real parallel subagents so their views are genuinely independent, not one context rationalizing with itself. REQUIRED: use [[dispatching-parallel-agents]] for the dispatch mechanics.
When to use
- Before committing to a plan, design, or launch where being wrong is expensive
- When you want critique, not encouragement
- For copy/landing pages: swap in the buyer-persona panel (below)
Not for generating options ([[brainstorming]]) or imagining failure of an already-chosen plan ([[premortem]]) — though they compose well.
How to run
- Gather the artifact — the idea/plan/draft to critique, plus any context the advisors need (goal, audience, constraints). Each subagent starts cold, so include everything in its prompt.
- Pick the panel. Default 5 (below). The user can override the count or the personas — honor that.
- Dispatch in parallel — one subagent per advisor, all in a single message. Give each:
- the artifact + context
- its persona brief and the one lens it must argue from
- the rule: be harsh, specific, and evidence-based; no hedging, no praise sandwich; cite the exact part you're attacking; return your top 3–5 issues, each with a severity (blocker / major / minor) and a concrete fix.
- Chairman pass (you, main context): collect all advisor outputs, dedupe overlaps, resolve disagreements (note where advisors conflict and why), rank by severity, and deliver a verdict.
Default panel
| Advisor | Lens |
|---|---|
| The Skeptic | Where's the flawed reasoning, the unsupported leap, the thing that won't survive contact with reality? |
| The User | Does this solve a real problem for the person it's for? Where will they get confused, bounce, or not care? |
| The Operator | Can this actually be built, shipped, and maintained? What's the hidden cost, the thing nobody owns? |
| The Red-teamer | How does this fail, get exploited, or get beaten by a competitor? Attack it. |
| The Domain Expert | Is it technically correct and rigorous? What does someone who knows this field deeply object to? |
Buyer-persona variant (for landing pages, emails, proposals): replace the panel with 5 target-customer personas reviewing the copy for conversion blockers — confusion, missing proof, weak CTA, price objections, trust gaps.
Output
- Verdict — ship / revise / kill, in one line
- Blockers then majors then minors — deduped and ranked across all advisors
- Conflicts — where advisors disagreed and your call on it
- Prioritized changes — the specific edits to make next
Common mistakes
- Letting advisors converge — give each a distinct lens and the explicit no-praise rule, or you get five copies of the same take.
- Skipping the chairman pass and dumping raw advisor output — the synthesis is the deliverable.
- Spawning advisors with too little context — they can't critique what they can't see.