advisor

Files (23)

• plugin.json · .claude-plugin/
• .python-version
• .standalone-db
• SKILL.md
• advisor.py
• advisor-roster.tsx · dashboard/components/
• decisions-board.tsx · dashboard/components/
• journal-list.tsx · dashboard/components/
• scenario-matrix.tsx · dashboard/components/
• lib.ts · dashboard/
• page.tsx · dashboard/pages/advisor/decision/[id]/
• page.tsx · dashboard/pages/advisor/
• route.ts · dashboard/routes/advisors/
• route.ts · dashboard/routes/decision/[id]/
• route.ts · dashboard/routes/decisions/
• route.ts · dashboard/routes/journal/
• decision.yaml · dashboard/views/
• hub.yaml · dashboard/views/
• hooks.json · hooks/
• pyproject.toml
• quality-checks.yaml
• schema.tql
• skill.yaml

SKILL.md (first lines)

# Advisor — Strategic Thought Partner / Board of Advisors

The 24/7 no-ego sounding board for the loneliness at the top: decisions you can't
yet take to the board or the team. A persona-based board of advisors debates the
question, then **converges**. Calibrated pushback — never sycophancy, never
devil's-advocate-for-sport. Bias surfacing, scenario stress-tests, and a decision
journal that closes the loop.

**Key principle:** the script stores and queries; Claude does the sensemaking —
writing each advisor's take in that seat's voice, running the debate, surfacing
biases, and drafting scenarios. The operator makes the call.

This skill obeys the repo's [operating principles](../../docs/operating-principles.md):
primer brain dumps at every open, interview-first, planning separated from
execution (the framing→debating→deciding states), and explicit
⏸ OPERATOR CHECKPOINT moments — everything between checkpoints is agent work.

---

## 1. Quick Start

```bash
# All commands: uv run --project skills/advisor python skills/advisor/advisor.py <command>

# Seat your first advisor
advisor.py add-advisor --name "The Operator" --archetype operator \
    --decision-style "execution-first" --pushback firm \
    --inspiration "Maria, my first COO" \
    --charter "Can we actually execute this with the team and cash we have?"

# Store personal context (do this once, keep it fresh)
advisor.py add-context --name "My role & mandate" --kind role \
    --content-file ~/context/role.md

# Open a decision with the primer brain dump
advisor.py open-decision \
    --question "Do we enter the EU market next quarter?" \
    --stakes high --operator-style pushback-then-space \
    --primer "Unfiltered thinking, exactly as spoken..."

# ... interview, takes, debate, bias check (see Decision Workflow) ...

# Record the call
advisor.py decide --decision decision-abc123 \
    --outcome "Enter EU, UK-first, 1 hire not 4" \
    --journal "What we decided, why, what we expect..." \
    --review-date 2026-10-01

# What's due for review?
advisor.py list-decisions --review-due
```

All commands emit a single JSON object on stdout; `report-decision` and
`report-board` emit Markdown for human display.

---

## 2. Building Your Board

A seat on the board is a **persona**, not a chatbot flavor. Build seats from:

- **Mentors you had** — the boss whose questions you still hear years later.
- **Mentors you wished for** — the seat you never had access to (a CFO who's
  seen three downturns, a founder who's sold a company).
- **Real thought-leaders as inspiration** — record who the seat channels in
  `--inspiration` ("thinks like Charlie Munger about incentives"). The
  inspiration anchors the voice; the **charter** defines the seat's job.

Give each seat:

| Field | What it does |
|---|---|
| `--archetype` | The seat's lens: operator, financier, contrarian, technologist, people-first, customer-champion... |
| `--decision-style` | How the persona reasons: first-principles, risk-first, pattern-matching from history, customer-back... |
| `--pushback` | `gentle` \| `firm` \| `relentless` — calibrated per seat |
| `--charter` | The standing question this seat always asks |

**Calibrating pushback.** This is the heart of the skill and the easiest thing
to get wrong in both directions:

- **Never sycophantic.** An advisor who agrees with the operator is dead weight;
  if a take amounts to "great idea, boss," rewrite it until it earns its seat.
- **Never devil's-advocate-for-sport.** Manufactured opposition that exists to
  perform rigor takes the wind out of good ideas and teaches the operator to
  stop bringing them. Pushback must be *in service of a better decision*.
- **Challenge, then CONVERGE.** Every debate ends with the board's converged
  recommendation (or a clearly-scoped disagreement the operator must resolve).
  A board that only argues has not done its job.

3–5 active seats is the sweet spot. Retire seats that stop earning their place
(`retire-advisor` — history is preserved, the seat leaves future debates).

⏸ OPERATOR CHECKPOINT — the operator names the seats, the inspirations, and the
pushback level for each. The agent drafts personas; the operator approves them.

---

## 3. Personal Context System

Context is everything. It is the difference between generic strategy advice and
an advisor who has worked with you for years.

Store context docs (`add-context --kind <kind>`):

| Kind | Contents |
|---|---|
| `role` | Your role, mandate, what you're accountable for, how you're measured |
| `company` | Stage, model, team shape, cash position, current constraints |
| `ecosystem` | Market, customers, partners, regulatory weather |
| `competitive-stance` | Who you're up against and how you win |
| `priorities` | This quarter's true priorities (not the official ones — the real ones) |
| `past-decision` | A previous decision and how it actually turned out |

Rules of use:

1. **Load context before every debate.** `list-context --full` and read the
   relevant docs before writing any take. Takes that ignore the operator's
   actual constraints are generic strategy content, not advice.
2. **Keep it fresh.** When the interview surfaces new context, offer to save it

From sciknow-io/alhazen-skill-personal-assistant, MIT.

More from sciknow-io/alhazen-skill-personal-assistant