Advisory board

Advisory Board (experimental — "Framework plus")

Convene a panel of advisor agents to review the team's worksheets from many angles at once, then synthesise where they agree, disagree, and what to test. This is a sparring partner, not an oracle — a mirror in your face, not a mirror of your own opinion: it exists to test the team's thinking, never to bless a decision already made. The board never assigns a score or picks the Primary. Every member is an AI lens grounded in a thinker's ideas, not the real person.

How to run it — detect the engine first

Step 0 — capability check. Before anything else, check whether a Workflow tool is available on this surface. The same entry point picks the engine from that one fact:

• IF Workflow is available (Claude Code) → run the bundled workflow (Path A).
• IF Workflow is NOT available (Cowork / Agent-SDK and other surfaces) → run the subagent-orchestrated board (Path B). Never block on the missing tool.

If the workspace is empty (no populated Worksheet 2), stop and advise the team to fill in Worksheet 2 first — the board reviews an existing analysis, it does not create one.

Path A — Workflow (Claude Code)

Convene the board via the bundled workflow (ships with the plugin — no install step). Pass a real object for args (the script also accepts a JSON-encoded string):

Workflow({
  scriptPath: "${CLAUDE_PLUGIN_ROOT}/workflows/wf-advisory-board.js",
  args: { worksheet: "<ws2 | ws3 | both>", workspace: "where-to-play", focus: "<optional factor/opportunity>" }
})

The workflow digests the worksheets, fans out every board member in parallel, and returns a synthesis (consensus tally / agreements / disagreements / top assumptions to test / blind spots). To prune the board (e.g. after the demo to the authors), pass args.board as an array of advisor agent ids, e.g. ["where-to-plai:advisor-sun-tzu", "where-to-plai:advisor-tal-gruber"].

Path B — subagent-orchestrated board (Cowork / Agent-SDK fallback)

When Workflow is unavailable, reproduce the same output contract by hand, in three phases:

a. Digest. Read the worksheet files in the workspace — where-to-play/02_attractiveness-map/ and where-to-play/02_map-overview.md for WS2; where-to-play/03_agile-focus.md for WS3 (read both for both). Produce a faithful, compact digest per opportunity: Overall Potential, Overall Challenge, zone, and key open assumptions (plus the Primary/Backup/Growth choices if WS3 is present). Do not score or editorialise — just summarise what the worksheets say. If no populated worksheet exists, stop and advise filling Worksheet 2 first.

b. Board. Spawn the bundled advisor agents as subagents via the Agent tool, using these subagent_type ids: where-to-plai:advisor-sun-tzu, where-to-plai:advisor-darwin, where-to-plai:advisor-socrates, where-to-plai:advisor-munger, where-to-plai:advisor-aristotle, where-to-plai:advisor-marcus-aurelius, where-to-plai:advisor-leonardo, where-to-plai:advisor-franklin, where-to-plai:advisor-moore, where-to-plai:advisor-christensen, where-to-plai:advisor-tal-gruber, plus where-to-plai:score-challenger. Give each subagent the digest and ask, through its own lens, for strengths, concerns, and assumptions to test. Prefer PARALLEL dispatch — several Agent calls in a single turn — where the surface supports it; sequential is acceptable otherwise. Each advisor runs in its own subagent context, and that independence is exactly what makes the consensus tally meaningful: a concern raised by many separate lenses carries weight because none of them saw the others' answers.

c. Synthesis. Yourself, summarise across the reviews — agreements; disagreements (name who diverges); top assumptions to test; blind spots (what no advisor raised but should be checked); and the consensus tally: merge semantically-equivalent concerns across advisors, count how many distinct advisors raised each, and sort descending. The tally is a weight-of-agreement signal on FINDINGS — never a score or a ranking of opportunities.

In both paths the board provokes and informs; it never scores, ranks, or picks a Primary — the team decides everything (human-in-the-loop). Personas are channeling, not impersonation; the feature is experimental and the roster can be pruned.

Portability note

Same entry point, two engines — the Step-0 branch picks one:

• Path A (Workflow / Claude Code): deterministic parallel fan-out with a schema-validated consensus tally.
• Path B (subagent / Cowork / SDK): the same output contract, with a model-computed tally.

After the board reports

1. Present the synthesis to the team: the consensus tally (which concerns/assumptions recur across the most advisors — a weight-of-agreement signal on findings, never a score or a ranking of opportunities), the agreements, the genuine disagreements (and who diverges), the top assumptions to test, and the blind spots.
2. Write the surfaced assumptions into where-to-play/assumptions.md so they enter the learning journey.
3. Hand back. Any change to a score or to the Primary/Backup/Growth choice is the team's call — route re-scoring to attractiveness-map and focus changes to agile-focus.

The board

See references/board.md for the roster, each advisor's lens, and the "channeling, not impersonation" framing. Default members: Sun Tzu, Darwin, Socrates, Munger, Aristotle, Marcus Aurelius, Leonardo, Franklin, Geoffrey Moore, Clayton Christensen, Tal & Gruber (duo), plus the score-challenger as the evidence lens.

Boundaries

• Experimental "Framework plus" feature — provocation, not verdict. The board's value is perspective, not a decision.
• Never let the board's output stand in for the team's judgement, and never present an advisor's view as the real person's actual opinion.