| name | ask-the-board |
| description | Convene your 3-person design board - "Elon Musk" (systems / architecture / data-model), "Steve Jobs" (taste & user experience), and "Tony Stark" (longevity & maintainability) - to spar with you on a new design or plan before you build. Each advisor reviews independently in parallel, then you get a synthesis of where they agree, where they clash, and the calls only you can make. Use when planning a new design or feature; invoke with /ask-the-board <what you are designing>. |
Ask the Board
Three sharp, opinionated advisors review your design in parallel - independent contexts so they do not converge - then you synthesize. This is your idea-sparring partner for the planning phase, before code exists.
1. Frame the proposal
Pull together what is actually being decided: the design / plan / feature in $ARGUMENTS plus relevant context from this conversation (and the open repo, if any). Write a tight 5-10 line PROPOSAL the advisors will react to - the goal, the user, the rough approach, and the open questions. If the ask is vague, sharpen it with one or two questions first; a fuzzy proposal yields fuzzy critique.
2. Convene the board - in parallel
Launch all three advisors at once, as independent sub-agents, so each gives a genuinely separate perspective instead of anchoring on the others (sequential review makes later advisors converge on the earlier ones):
advisor-elon-musk - first principles; system, architecture, data model; delete the requirement; survive 10x scale.advisor-steve-jobs - taste and user experience; prototypes the user-facing shell first; searches the web and Reddit for the best current UX; loves Claude Design.advisor-tony-stark - longevity and maintainability; what becomes a maintenance tax at scale.
Give each the same PROPOSAL and ask for its structured verdict. Run them as parallel Agent calls (subagent_type = each advisor name) in a single message.
3. Synthesize the board meeting
Once all three report, write the synthesis - do NOT just paste three reviews:
- Consensus - what all three agree on; treat as settled, do it.
- Tensions - where they clash, framed as the real decision. The signature clash to expect and name: Jobs ("prototype the shell first, validate the experience") vs Musk ("get the data model and architecture right first") - with Stark weighing how each path ages. Name it; do not paper over it.
- Top risks - the few things most likely to bite, across all three lenses.
- Recommended path - your read of the best route given the three verdicts, with the trade-off stated plainly.
- Decisions for you - the 1-3 calls only the user can make.
Keep the advisors in character, but keep the synthesis decision-focused and specific to the design under review.
Notes
- Advisors are read-only: they critique, they do not touch code.
- New advisor agents load at session start. If
/ask-the-board cannot find them, restart the session once after they are created.
- Add, swap, or retune advisors by editing
~/.claude/agents/advisor-*.md; keep this skill's roster in sync (and update SETUP.html / TEMPLATE.md per the workspace template convention).
Gotchas
- Do not let the board converge - launch the three in parallel and keep their reviews independent; sequential review makes later advisors anchor on earlier ones. (learned 2026-06-16)
