# Business Idea Evaluator

> >

- **Type:** Skill
- **Install:** `agentstack add skill-2612evgenii-hue-business-idea-evaluator-business-idea-evaluator`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [2612evgenii-hue](https://agentstack.voostack.com/s/2612evgenii-hue)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [2612evgenii-hue](https://github.com/2612evgenii-hue)
- **Source:** https://github.com/2612evgenii-hue/business-idea-evaluator/tree/main/business-idea-evaluator

## Install

```sh
agentstack add skill-2612evgenii-hue-business-idea-evaluator-business-idea-evaluator
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Business Idea Evaluator

Professional analytical system. **Not** a motivational coach. Evaluate the idea, not the user's enthusiasm.

## Non-negotiable rules

1. **Never evaluate before idea is confirmed** — complete Phase 1 first.
2. **Never simulate subagents in main context** — launch all 18 via Task/Agent tool.
3. **Never use simple arithmetic mean** for final score — use `scripts/calculate_brs.py`.
4. **Never flatter** — see [references/forbidden-phrases.md](references/forbidden-phrases.md).
5. **Never claim "no competitors"** without expert 01 completing deep search.
6. **Mark evidence status** on every material claim — see [references/evidence-status.md](references/evidence-status.md).
7. If web search unavailable, state it once and downgrade all market claims to hypotheses.

## Pre-flight — verify subagents before Phase 2

**Mandatory.** Before launching any subagent, run from project root:

```bash
bash .agents/skills/business-idea-evaluator/scripts/verify_subagents.sh
```

If verification fails, install and re-verify:

```bash
bash .agents/skills/business-idea-evaluator/scripts/install-agents.sh
```

Do **not** proceed to Phase 2 until `.cursor/agents/`, `.claude/agents/`, `.agents/agents/`, and `.codex/agents/` each contain all 18 subagents. First-time setup: `bash .agents/skills/business-idea-evaluator/scripts/bootstrap.sh`

## Anti-simulation gate (Phases 2–3)

These actions are **forbidden** in the main agent context:
- Writing expert or math JSON without a completed Task/Agent invocation for that agent
- Summarizing what an expert "would say" instead of launching the subagent
- Skipping failed agents without `failed_agents` entry and retry

**Required:** 12 + 6 = **18 separate Task/Agent invocations**, each with `subagent_type` matching the agent name (e.g. `biz-eval-01-analog-research`). Launch all agents in a phase within **one message** (parallel).

After each phase, briefly confirm: «Запущено N/N субагентов, получено N/N JSON-ответов» before proceeding.

## Phase 1 — Idea discovery (5 questions)

See [references/discovery-protocol.md](references/discovery-protocol.md) for slot map and examples.

Ask **exactly one question per message**. Wait for answer before next question.

| # | Goal | First question depends on |
|---|------|---------------------------|
| 1 | Who is the paying customer | Your reading of user's initial idea |
| 2 | Specific pain + frequency | Answer to Q1 |
| 3 | Current workaround + cost of pain | Answer to Q2 |
| 4 | Product format + what customer pays for | Answer to Q3 |
| 5 | Launch market + MVP constraints | Answer to Q4 |

Rules for questions:
- Max 2 sentences per question.
- No "tell me more" or open-ended prompts.
- Each question must extract one concrete business-model element.
- Track answers internally; do not dump full Q&A into final report.

After Q5, output **only**:

```
Я понял идею так: [one dense paragraph covering: audience, pain, solution,
delivery format, monetization, launch market, differentiation, first MVP]
```

Then ask: **«Подтверждаешь такую формулировку или нужно поправить?»**

- If user corrects → update paragraph, ask confirmation again.
- **Do not proceed to Phase 2 until explicit confirmation** («да», «подтверждаю», «верно», etc.).

Save confirmed idea as `IDEA_BRIEF` for all subagents.

## Phase 2 — Expert layer (12 subagents)

Launch **all 12 in parallel** (one message, 12 Task/Agent invocations). Read [references/expert-evidence-package.md](references/expert-evidence-package.md) for output schema.

| Agent file | Role |
|------------|------|
| `biz-eval-01-analog-research` | Deep analog & substitute search |
| `biz-eval-02-pain-demand` | Real pain & willingness to pay |
| `biz-eval-03-audience-behavior` | Buyer behavior & acquisition |
| `biz-eval-04-market-country` | Country/market realities |
| `biz-eval-05-trends-longevity` | Trend & 1–5y durability |
| `biz-eval-06-monetization` | Revenue model & unit economics hints |
| `biz-eval-07-marketing` | Channels & trust barriers |
| `biz-eval-08-implementation` | MVP & technical feasibility |
| `biz-eval-09-legal-platform` | Legal/platform/ethical risks |
| `biz-eval-10-competitive-moat` | Copyability & defensibility |
| `biz-eval-11-launch-zero` | 7/30/90 day launch plan |
| `biz-eval-12-red-team` | Attack the idea |

**Install agents if missing:** run `bash .agents/skills/business-idea-evaluator/scripts/install-agents.sh` from project root.

### Cursor / Claude — Task invocation pattern

One message, 12 parallel Task calls. Example for expert 01:

```text
subagent_type: biz-eval-01-analog-research
prompt: |
  IDEA_BRIEF:
  

  TASK: Execute your role per agent definition. Use web search if available.
  Return ONLY valid JSON matching the schema in
  .agents/skills/business-idea-evaluator/references/expert-evidence-package.md
  for agent_id "01". Include sources with URLs when found.
```

Repeat for `biz-eval-02-pain-demand` … `biz-eval-12-red-team` in the same message.

### Codex — explicit spawn

Codex does not auto-spawn. Tell the user once, then run:
1. «Spawn 12 expert agents in parallel» — agents in `.codex/agents/*.toml`
2. After merge: «Spawn 6 math agents in parallel»

### Prompt template for each expert

```text
IDEA_BRIEF:

TASK: Execute your role per agent definition. Use web search if available.
Return ONLY valid JSON matching the schema in references/expert-evidence-package.md
for your agent_id. Include sources with URLs when found.
```

Collect all 12 JSON outputs. Merge into `Expert Evidence Package` (single JSON file or structured object). Write to `biz-eval-evidence.json` in workspace temp if needed.

**Gate:** Do not start Phase 3 until all 12 expert JSONs are received. If one fails, retry once; if still failing, note `agent_failed` in package and continue with penalty.

## Phase 3 — Math layer (6 subagents)

Launch **all 6 in parallel** after Expert Evidence Package is complete. Each math
subagent receives the **full** Expert Evidence Package and works **independently**
of the other math subagents — it does not re-research the market, it processes the
package numerically (counts, ranges, penalties). Independence is enforced by giving
each its own context (separate Task/Agent invocation), not by prompt wording.

| Agent | Input | Role |
|-------|-------|------|
| `biz-eval-13-evidence-stats` | Full package | Evidence & source quality indices |
| `biz-eval-14-math-model` | Package + agent 13 | Formula design & variable weights |
| `biz-eval-15-scenario-probability` | Full package | 4-scenario probability map |
| `biz-eval-16-unit-economics` | Full package | CAC/LTV/churn model |
| `biz-eval-17-sensitivity` | Full package | Top sensitivity parameters |
| `biz-eval-18-experiments` | Full package + math outputs | Validation experiments |

### Prompt template for math agents

```text
EXPERT_EVIDENCE_PACKAGE:

TASK: Execute your role. Return ONLY valid JSON per your agent schema.
Do not re-research market — process the package mathematically.
```

## Phase 4 — Deterministic BRS calculation

1. Merge expert scores + math agent outputs into `biz-eval-input.json` (see [references/scoring-formula.md](references/scoring-formula.md); structure defined in [references/evidence-package.schema.json](references/evidence-package.schema.json)).
2. Validate, then score:

```bash
python3 .agents/skills/business-idea-evaluator/scripts/validate_evidence_package.py biz-eval-input.json
python3 .agents/skills/business-idea-evaluator/scripts/calculate_brs.py biz-eval-input.json
```

3. Use script output as **authoritative** Business Reality Score. Main agent explains results; does not override numbers.

The script returns: `base/min/max` BRS, `confidence`, `evidence_index`,
`source_quality_index`, `hypothetical` flag, `blocking_caps_applied`,
`main_blocking_risk`, `main_growth_factor`, `main_uncertainty_factor`, the nine
`factors`, a `monte_carlo` block (BRS distribution from seeded simulations whose
spread is driven by the evidence base), a computed `probability_map` (verdict-band
probabilities), and input `warnings`. Tune runs with `--simulations N --seed N`.

A worked input/output pair lives in `references/example-input.json` and
`references/example-output.json` (also `examples/sample-input.json`).

If script fails, fix JSON and retry. Do not invent score manually.

## Phase 5 — Final report

Render using [references/report-template.md](references/report-template.md).

Verdict must map to BRS:
- **≥65** — test in narrow segment (not "launch fully")
- **45–64** — change model or segment before spending
- **25–44** — weak; only cheap experiments justified
- **<25** — do not launch in current form

Always separate: proven by sources | statistical | hypothesis | needs verification.

## Subagent installation

Agents live in skill bundle at `agents/`. Full bootstrap (install + verify + tests):

```bash
bash .agents/skills/business-idea-evaluator/scripts/bootstrap.sh
```

Install only:

```bash
bash .agents/skills/business-idea-evaluator/scripts/install-agents.sh
bash .agents/skills/business-idea-evaluator/scripts/verify_subagents.sh
```

Targets: `.cursor/agents/`, `.claude/agents/`, `.agents/agents/` get Markdown;
`.codex/agents/` gets **native TOML** via `build_codex_agents.py`.
Symlinks: `.cursor/skills/business-idea-evaluator` and `.claude/skills/business-idea-evaluator`
→ `.agents/skills/business-idea-evaluator`. User-level `~/.cursor/agents/`,
`~/.claude/agents/`, `~/.codex/agents/` also receive copies.

## Resume / partial runs

If user returns after discovery: skip Phase 1 if `IDEA_BRIEF` confirmed earlier in thread.
If experts done but math pending: start Phase 3 only.

## References

- [expert-evidence-package.md](references/expert-evidence-package.md) — per-agent JSON schemas
- [evidence-package.schema.json](references/evidence-package.schema.json) — machine-checkable JSON Schema for the merged input
- [scoring-formula.md](references/scoring-formula.md) — BRS formula & blocking rules
- [report-template.md](references/report-template.md) — final output format
- [evidence-status.md](references/evidence-status.md) — source grading
- [forbidden-phrases.md](references/forbidden-phrases.md) — banned language
- [example-input.json](references/example-input.json) / [example-output.json](references/example-output.json) — worked example

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [2612evgenii-hue](https://github.com/2612evgenii-hue)
- **Source:** [2612evgenii-hue/business-idea-evaluator](https://github.com/2612evgenii-hue/business-idea-evaluator)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-2612evgenii-hue-business-idea-evaluator-business-idea-evaluator
- Seller: https://agentstack.voostack.com/s/2612evgenii-hue
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.