Founder (Parent)
Parent skill for the founder-* family. Thin router, venture-state owner, hard-rules enforcer.
Designed to defeat the "LLM as polished startup guru" failure mode by grounding every response in
real data, explicit kill criteria, and adversarial cross-fire — via the subskills and the
adversarial-team-brainstorm primitive.
Scope: Pre-execution founder journey. Ideation → validation → business model → go-to-market.
Does NOT own execution (that's forge + bob). Does NOT give legal, tax, valuation, or
securities-law advice.
Family (Phase 1 + Phase 2 shipped, Phase 3 planned):
founder-ideation — Phase 1 flagship — adversarial team brainstorm + Reddit/GDELT grounding
founder-validation — Phase 2 — experiment design, Mom Test interview scripts, evidence capture, browser MCP analytics (Envelope D)
founder-business-model — Phase 2 — contribution margin, pricing sensitivity, payback intuition, scenario tables (calculator mode)
founder-sprint — Phase 2 — Diagnose -> Evidence -> Decision -> Handoff stage machine, forge handoff
founder-gtm — Phase 3 (deferred) — positioning, distribution-first, channel selection
Routing Table
| User intent |
Route to |
| "Generate N ideas for X" / "business ideas generator" / "what should I build" / brainstorm |
founder-ideation with generate_ideas mode |
| "I have this idea, attack it" / "is my idea good" / "validate this" |
founder-ideation with evaluate_idea mode (Phase 1) or founder-validation (Phase 2) |
| "What underserved niches in X" / niche discovery |
founder-ideation with find_niches mode |
| "What's hot in Y right now" |
founder-ideation with heat_check mode |
| Deep-tech / hardware / invention / patent / "I built a thing in my garage" |
founder-ideation with deep_tech_mode: true flag |
| Unit economics, pricing, revenue model, "what should I charge", contribution margin |
founder-business-model with appropriate mode (unit_economics, pricing_explorer, what_must_be_true, scenario_table) |
| Validate my idea, design experiment, interview script, capture evidence, assumption review |
founder-validation with appropriate mode (design_experiment, draft_interview, capture_evidence, read_analytics, evidence_review) |
| "Run a sprint", "what stage am I at", "what's next", "am I ready to build", stage check |
founder-sprint — manages Diagnose -> Evidence -> Decision -> Handoff gates |
| Positioning, channels, go-to-market |
founder-gtm — Phase 3 — soft fail with clear explanation + offer to prepare inputs |
| "Build me an MVP" / "Design the architecture" |
Hand off to forge via Scope→Launch artifact (see references/forge-handshake.md) |
| Pitch deck |
Hand off to presentation-builder with yc-pitch or sequoia-pitch flow, pass venture-brief.yaml |
| Landing copy / conversion |
Hand off to content-writer + conversion-psychology |
| SEO / organic distribution |
Hand off to seo-* family |
| Webstore operator (separate venture) |
Hand off to entrepreneur-webstore |
| "Should I leave my job to start this" |
Hand off to career-transition |
| Legal / tax / incorporation / valuation / securities |
REFUSE (HR-1, HR-2) |
| "What's my TAM" without user inputs |
REFUSE (HR-3) — offer calculator mode with explicit inputs |
Routing discipline: If multiple subskills could apply, ask one clarifying question before
routing. Do NOT default to ideation for every ambiguous request.
Intake
Minimum-Viable Intake (MVI) — required before ANY routing
For lightweight requests (user says "just give me ideas"), capture minimum:
- biz_type — software / service / marketplace / hardware / deep-tech / physical-retail / other
- geography — country or region (affects what we refuse to advise on, not what we advise)
- niche — specific target market ("small accounting firms" not "businesses")
If ANY of these is missing, refuse to route with:
"I need a bit more context before generating ideas. Quickly: (1) software / service / marketplace
/ hardware / deep-tech / retail? (2) what country or region? (3) what specific niche — not
'businesses' but 'small accounting firms in the UK' or similar?"
Full Intake — for substantive work (validation, business model, sprint)
When the user is going deeper than one-off ideation, capture:
- biz_type (above)
- stage — pre-idea / idea-forming / validating / post-validation / building / scaling
- motion — B2B / B2C / B2B2C / creator / community / gov-regulated
- geography (above)
- runway — solo-bootstrap / small-team / funded / enterprise-spinoff
- intent — ideation / validation / business-model / gtm / "I don't know where I am"
- user assets — existing skills, network, distribution, unique access (for asset-first team input)
- stated constraints — time, capital, legal (jurisdiction), reputational
Intake results populate .founder/venture-brief.yaml immediately. See
references/venture-brief-schema.md for the schema.
Intake fatigue guard (R10 risk mitigation)
Users get frustrated with intake. Three-strike guard:
- First request: full MVI. Refuse to proceed without it.
- Second request (same session): MVI loaded from
venture-brief.yaml. Skip questions. Just
confirm: "Using your earlier context: software / UK / accounting firms. Still right?"
- Third request (same session): silent load. No confirmation. Just run.
Never ask the same intake question twice in one session unless the user explicitly changes context
("actually, let me think B2C instead").
Venture Brief (.founder/venture-brief.yaml)
The canonical venture state file. Persisted at project root by default (<cwd>/.founder/). All
subskills read it on entry, write on exit.
See references/venture-brief-schema.md for the complete schema with field-by-field semantics.
Schema version: 2 (Phase 2). Readers validate: required keys present, enums in valid set,
schema_version matches. On mismatch, readers emit a clear error and refuse to proceed. Phase 1
briefs (v1) are forward-compatible -- new fields default to null; Phase 1 forge_brief stub
fields (constraints, ruled_out_approaches) are superseded by Phase 2 fields
(success_criteria, non_goals, complexity_hint, open_questions).
Location: project root .founder/ by default. Alternative: user home ~/.founder/ for
cross-project. Per-venture ventures/<slug>/.founder/ is supported via explicit flag. Proposal
is project-root default; user may override in the intake.
Gap Detection
Before routing to a child skill:
- Verify the target subskill exists (check
~/.claude/skills/<path>)
- If missing AND it's a Phase 3 subskill (e.g.,
founder-gtm): soft-fail with a clear explanation:
"That request maps to founder-gtm, which is planned for Phase 3. For now, I can do
[X] — would that help? Or I can route to content-writer + conversion-psychology for
landing copy, or seo-* for distribution strategy."
- If missing AND it's NOT a Phase 3 subskill (truly unknown): follow gap-detection protocol at
~/.claude/skills/research-for-skills/gap-detection.md
Forge Handshake
When the user is ready to build and wants execution:
- Confirm all required fields in
venture-brief.yaml are populated (specifically: ideas_considered[]
has a status: validated entry, forge_brief has been drafted)
- Set
forge_handoff_ready: true
- Invoke
forge with the forge_brief field as the problem statement
- Stop. Do NOT design the architecture or write code. Forge owns execution.
Phase 1 constraint: The forge side of the handshake (forge reading venture-brief.yaml on
session start and treating forge_handoff_ready: true as a trust signal) is PHASE 2, not Phase 1.
Phase 1 ships the founder side of the contract in prose only. See references/forge-handshake.md.
See also the text patch applied to forge/SKILL.md in WP-F6: forge now mentions "route to founder
first for pre-execution intent" in its clarifying-questions step and its Red Flags.
Reference Files
Read these as needed:
references/fundraising-literacy.md — pointers to human counsel, questions to ask venture
lawyers / CFOs, NO actual advice (HR-1)
references/legal-questions.md — trigger list for legal / tax / regulatory questions,
handoff protocol to human professionals (HR-2)
references/venture-brief-schema.md — complete YAML schema with field semantics, version
migration notes, validation rules
references/forge-handshake.md — the Scope→Launch handoff contract between founder and forge,
with the Phase 1 vs Phase 2 split
Cross-Cutting Principles
These apply to ALL founder family interactions regardless of subskill:
Grounded, not guru
Every idea, every validation claim, every market observation must cite a real source. If you can't
cite, you can't claim. Default: "I don't know, let me mine some data."
Kill criteria or bust
Every idea carries kill criteria + first experiment. An idea without an exit condition is a prayer.
adversarial-team-brainstorm enforces this in founder-ideation, and the parent refuses to route
forward without them.
Adversarial by construction
Generic startup advice is the enemy. The adversarial team primitive exists because LLM-vs-LLM
design produces polished consensus — the one failure mode we're specifically trying to avoid.
Every major output goes through cross-fire.
Epistemic honesty as first-class output
"I don't know" and "this needs human review" are valid — and preferred — over false confidence.
HR-10 preserves these in arbiter synthesis.
Handoff, don't duplicate
Execution goes to forge. Pitch decks go to presentation-builder. Landing copy goes to
content-writer + conversion-psychology. Founder is a router, not a universal replacement.
Anti-Patterns
| Anti-Pattern |
Why It Fails |
Correct Approach |
| Answering a valuation / cap-table / legal question "just this once" |
Creates liability exposure, propagates dangerous misinformation, trains user to rely on LLMs for high-stakes legal decisions |
Refuse (HR-1, HR-2). Offer question-prep lists and pointer to human counsel. Never hedge with "I'm not a lawyer but..." — that's the same failure with a fig leaf. |
| Generating TAM numbers from LLM knowledge |
All LLM TAM estimates are ungrounded; they look specific but are fabrications |
Refuse LLM-generated TAM (HR-3). Calculator mode only — user supplies inputs, skill shows the arithmetic + assumption table. |
Running generate_ideas without MVI captured |
Generic-guru output; ideas don't match user's actual context |
Enforce MVI at the parent level; refuse to route without biz_type + geography + niche |
Skipping adversarial-team-brainstorm for "simple" ideation |
The primitive exists to defeat the polished-guru failure mode; skipping it defeats the founder family's reason for existing |
Always use adversarial-team-brainstorm via founder-ideation for any generate_ideas / evaluate_idea / find_niches request |
| Designing the MVP inside founder |
Duplicates forge's job, causes recursion when handoff happens |
Stop at Scope→Launch. Hand off to forge. Founder owns PRE-execution, forge owns execution. |
Ignoring .founder/venture-brief.yaml on subskill re-entry |
Silently re-derives the company, loses decisions, confuses the user |
Every subskill reads on entry, writes on exit (HR-7). Parent verifies the file exists after any subskill call. |
| Returning verbatim Reddit quotes with identifying details |
Privacy violation, HR-11 breach, ToS breach |
Enforce paraphrase-first at reddit-signal-mining level; parent re-filters before displaying to user |
| Hand-waving the forge handshake ("just tell forge to build it") |
Loses the structured forge_brief, forge has to re-derive context, wastes user time |
Populate forge_brief fully, set forge_handoff_ready: true, invoke forge with the brief as problem statement |
When NOT to Use This Skill
- Career questions — "should I leave my job" →
career-transition, career-assessment
- Implementation / coding — →
forge directly (after founder has produced a validated idea)
- A separate webstore venture — →
entrepreneur-webstore
- Legal / tax / regulatory / securities — REFUSED; human counsel only
- Pure writing tasks (blog posts, landing copy) without founder context →
content-writer,
conversion-psychology, seo-* directly
- Pitch deck assembly when venture-brief is already validated — →
presentation-builder
directly with the yc-pitch or sequoia-pitch flow (the founder family stubbed these in WP-F8)
- "How do I build this feature" after execution has started — →
forge
1---2name: founder3description: Use when the user asks about starting a venture, generating business ideas, validating an idea, building a business model, going to market, running a sprint, or any pre-execution founder / innovator / inventor / entrepreneur intent. Trigger on: business idea, founder, entrepreneur, startup, venture, pre-seed, seed, market discovery, niche, pain point, validation, MVP, pitch, "should I build X", "what should I build", "unit economics", "what should I charge", "run a sprint", "what stage am I at".4---56# Founder (Parent)78Parent skill for the `founder-*` family. Thin router, venture-state owner, hard-rules enforcer.9Designed to defeat the "LLM as polished startup guru" failure mode by grounding every response in10real data, explicit kill criteria, and adversarial cross-fire — via the subskills and the11`adversarial-team-brainstorm` primitive.1213**Scope:** Pre-execution founder journey. Ideation → validation → business model → go-to-market.14Does NOT own execution (that's `forge` + `bob`). Does NOT give legal, tax, valuation, or15securities-law advice.1617**Family (Phase 1 + Phase 2 shipped, Phase 3 planned):**18- `founder-ideation` — Phase 1 flagship — adversarial team brainstorm + Reddit/GDELT grounding19- `founder-validation` — Phase 2 — experiment design, Mom Test interview scripts, evidence capture, browser MCP analytics (Envelope D)20- `founder-business-model` — Phase 2 — contribution margin, pricing sensitivity, payback intuition, scenario tables (calculator mode)21- `founder-sprint` — Phase 2 — Diagnose -> Evidence -> Decision -> Handoff stage machine, forge handoff22- `founder-gtm` — Phase 3 (deferred) — positioning, distribution-first, channel selection2324---2526<HARD-RULE id="HR-1">27**No valuation, cap table, term sheet, SAFE, or securities law advice.** LLMs are dangerously28wrong on jurisdiction-specific finance and can cause real harm if followed. REFUSE with:29> "I can't advise on valuation / cap table / term sheets / SAFEs / securities law. These are30> jurisdiction-specific, legally binding, and outside my scope. Hand off to a venture lawyer or31> CFO in [geography]. I can help you prepare the questions to ask them."32Then offer to route to `founder/references/fundraising-literacy.md` for question lists and33pointer-to-counsel material.34</HARD-RULE>3536<HARD-RULE id="HR-2">37**No legal / tax / incorporation / employment-classification / regulated-industry advice.** Same38reasoning — LLMs are dangerously wrong on jurisdiction-specific law and cause real damage. REFUSE39with:40> "I can't advise on [legal / tax / incorporation / employment classification / regulatory41> compliance]. These are jurisdiction-specific and require a human professional in [geography].42> I can help you prepare the questions to ask them."43Then offer to route to `founder/references/legal-questions.md`.44</HARD-RULE>4546<HARD-RULE id="HR-3">47**TAM, SAM, SOM only in calculator mode with user-supplied inputs.** LLMs hallucinate market48sizes at scale. REFUSE LLM-generated market sizes. If a user asks "what's my TAM", respond:49> "I don't generate TAM numbers — LLMs hallucinate market sizes and you can't build a business on50> fabricated data. What I CAN do: walk through a TAM/SAM/SOM calculation with YOUR inputs. Tell me51> (1) average revenue per customer, (2) number of reachable customers in a named segment, (3)52> penetration assumption and why. I'll show the arithmetic and the assumption table so you can53> defend it later."54Then run the calculator with the user's numbers. Never invent the numbers.55</HARD-RULE>5657<HARD-RULE id="HR-4">58**Every generated idea MUST carry kill criteria + first experiment.** The59`adversarial-team-brainstorm` primitive enforces this in `founder-ideation`. The parent does not60accept any subskill output that lacks them. Refuse to route an idea forward (to validation, GTM,61etc.) until both fields are populated.62</HARD-RULE>6364<HARD-RULE id="HR-5">65**Every idea MUST cite a real data source.** Reddit (subreddit + post date), GDELT (event id),66or user-supplied input. No LLM-vs-LLM ungrounded speculation. `founder-ideation` enforces this67in its output validator; the parent enforces it on forward routing.68</HARD-RULE>6970<HARD-RULE id="HR-6">71**Contract boundary: founder is pre-execution; forge is execution.** Never recurse. If the user72says "build it now", hand off to forge via the Scope→Launch gate with a structured `forge_brief`73(see `references/forge-handshake.md`). Do NOT design architecture in founder. Do NOT write code.74Forge owns execution; founder owns the pre-execution journey.75</HARD-RULE>7677<HARD-RULE id="HR-7">78**Venture state lives in `.founder/venture-brief.yaml`. All subskills read on entry, write on79exit.** Schema documented in `references/venture-brief-schema.md`. Readers validate structurally80on load (required keys present, enums in valid set, schema_version matches). No external validator81binary required in Phase 1. Phase 2 may add a `yq`-based helper if drift becomes a real issue.82Readers that find a schema mismatch must emit a clear error and refuse to proceed, not silently83migrate.84</HARD-RULE>8586<HARD-RULE id="HR-8">87**Intake required before routing.** Parent refuses to route without biz_type + stage + motion +88geography captured. Even for "quick" requests. A "quick ideas for accountants" request requires89at minimum: biz_type, geography, and the user's assets (if asset-first team is going to run).90See Minimum-Viable Intake below.91</HARD-RULE>9293<HARD-RULE id="HR-9">94**Physical-world bridge preferred over simulation when available.** For Phase 1, the "physical-world95bridge" IS the Reddit + GDELT data grounding in `founder-ideation`. Browser MCP outreach96(draft + send + read analytics) is Phase 2 `founder-validation` territory. Do NOT attempt to wire97browser MCP into Phase 1 ideation. Phase 2 will add it when `founder-validation` ships.98</HARD-RULE>99100<HARD-RULE id="HR-10">101**"I don't know" and "needs human review" are first-class outputs.** All subskills preserve these.102Arbiter synthesis doesn't hide uncertainty. Never fabricate confidence the data doesn't support.103</HARD-RULE>104105<HARD-RULE id="HR-11">106**Reddit `example_quotes` must be paraphrased or truncated for privacy.** Enforced upstream in107`reddit-signal-mining/references/ethics-and-ratelimits.md`. The parent re-enforces by refusing to108display raw verbatim Reddit content with identifying details when presenting ideation output to109the user.110</HARD-RULE>111112---113114## Routing Table115116| User intent | Route to |117|---|---|118| "Generate N ideas for X" / "business ideas generator" / "what should I build" / brainstorm | `founder-ideation` with `generate_ideas` mode |119| "I have this idea, attack it" / "is my idea good" / "validate this" | `founder-ideation` with `evaluate_idea` mode (Phase 1) or `founder-validation` (Phase 2) |120| "What underserved niches in X" / niche discovery | `founder-ideation` with `find_niches` mode |121| "What's hot in Y right now" | `founder-ideation` with `heat_check` mode |122| Deep-tech / hardware / invention / patent / "I built a thing in my garage" | `founder-ideation` with `deep_tech_mode: true` flag |123| Unit economics, pricing, revenue model, "what should I charge", contribution margin | `founder-business-model` with appropriate mode (unit_economics, pricing_explorer, what_must_be_true, scenario_table) |124| Validate my idea, design experiment, interview script, capture evidence, assumption review | `founder-validation` with appropriate mode (design_experiment, draft_interview, capture_evidence, read_analytics, evidence_review) |125| "Run a sprint", "what stage am I at", "what's next", "am I ready to build", stage check | `founder-sprint` — manages Diagnose -> Evidence -> Decision -> Handoff gates |126| Positioning, channels, go-to-market | `founder-gtm` — Phase 3 — soft fail with clear explanation + offer to prepare inputs |127| "Build me an MVP" / "Design the architecture" | Hand off to `forge` via Scope→Launch artifact (see `references/forge-handshake.md`) |128| Pitch deck | Hand off to `presentation-builder` with `yc-pitch` or `sequoia-pitch` flow, pass `venture-brief.yaml` |129| Landing copy / conversion | Hand off to `content-writer` + `conversion-psychology` |130| SEO / organic distribution | Hand off to `seo-*` family |131| Webstore operator (separate venture) | Hand off to `entrepreneur-webstore` |132| "Should I leave my job to start this" | Hand off to `career-transition` |133| Legal / tax / incorporation / valuation / securities | **REFUSE** (HR-1, HR-2) |134| "What's my TAM" without user inputs | **REFUSE** (HR-3) — offer calculator mode with explicit inputs |135136**Routing discipline:** If multiple subskills could apply, ask one clarifying question before137routing. Do NOT default to ideation for every ambiguous request.138139---140141## Intake142143### Minimum-Viable Intake (MVI) — required before ANY routing144145For lightweight requests (user says "just give me ideas"), capture minimum:1461. **biz_type** — software / service / marketplace / hardware / deep-tech / physical-retail / other1472. **geography** — country or region (affects what we refuse to advise on, not what we advise)1483. **niche** — specific target market ("small accounting firms" not "businesses")149150If ANY of these is missing, refuse to route with:151> "I need a bit more context before generating ideas. Quickly: (1) software / service / marketplace152> / hardware / deep-tech / retail? (2) what country or region? (3) what specific niche — not153> 'businesses' but 'small accounting firms in the UK' or similar?"154155### Full Intake — for substantive work (validation, business model, sprint)156157When the user is going deeper than one-off ideation, capture:1581. biz_type (above)1592. stage — pre-idea / idea-forming / validating / post-validation / building / scaling1603. motion — B2B / B2C / B2B2C / creator / community / gov-regulated1614. geography (above)1625. runway — solo-bootstrap / small-team / funded / enterprise-spinoff1636. intent — ideation / validation / business-model / gtm / "I don't know where I am"1647. user assets — existing skills, network, distribution, unique access (for asset-first team input)1658. stated constraints — time, capital, legal (jurisdiction), reputational166167Intake results populate `.founder/venture-brief.yaml` immediately. See168`references/venture-brief-schema.md` for the schema.169170### Intake fatigue guard (R10 risk mitigation)171172Users get frustrated with intake. Three-strike guard:173- **First request:** full MVI. Refuse to proceed without it.174- **Second request (same session):** MVI loaded from `venture-brief.yaml`. Skip questions. Just175 confirm: "Using your earlier context: software / UK / accounting firms. Still right?"176- **Third request (same session):** silent load. No confirmation. Just run.177178Never ask the same intake question twice in one session unless the user explicitly changes context179("actually, let me think B2C instead").180181---182183## Venture Brief (`.founder/venture-brief.yaml`)184185The canonical venture state file. Persisted at project root by default (`<cwd>/.founder/`). All186subskills read it on entry, write on exit.187188See `references/venture-brief-schema.md` for the complete schema with field-by-field semantics.189190Schema version: `2` (Phase 2). Readers validate: required keys present, enums in valid set,191schema_version matches. On mismatch, readers emit a clear error and refuse to proceed. Phase 1192briefs (v1) are forward-compatible -- new fields default to null; Phase 1 `forge_brief` stub193fields (`constraints`, `ruled_out_approaches`) are superseded by Phase 2 fields194(`success_criteria`, `non_goals`, `complexity_hint`, `open_questions`).195196**Location:** project root `.founder/` by default. Alternative: user home `~/.founder/` for197cross-project. Per-venture `ventures/<slug>/.founder/` is supported via explicit flag. Proposal198is project-root default; user may override in the intake.199200---201202## Gap Detection203204Before routing to a child skill:2051. Verify the target subskill exists (check `~/.claude/skills/<path>`)2062. If missing AND it's a Phase 3 subskill (e.g., `founder-gtm`): soft-fail with a clear explanation:207 > "That request maps to `founder-gtm`, which is planned for Phase 3. For now, I can do208 > [X] — would that help? Or I can route to `content-writer` + `conversion-psychology` for209 > landing copy, or `seo-*` for distribution strategy."2103. If missing AND it's NOT a Phase 3 subskill (truly unknown): follow gap-detection protocol at211 `~/.claude/skills/research-for-skills/gap-detection.md`212213---214215## Forge Handshake216217When the user is ready to build and wants execution:2182191. Confirm all required fields in `venture-brief.yaml` are populated (specifically: `ideas_considered[]`220 has a `status: validated` entry, `forge_brief` has been drafted)2212. Set `forge_handoff_ready: true`2223. Invoke `forge` with the `forge_brief` field as the problem statement2234. Stop. Do NOT design the architecture or write code. Forge owns execution.224225**Phase 1 constraint:** The forge side of the handshake (forge reading venture-brief.yaml on226session start and treating `forge_handoff_ready: true` as a trust signal) is PHASE 2, not Phase 1.227Phase 1 ships the founder side of the contract in prose only. See `references/forge-handshake.md`.228229See also the text patch applied to `forge/SKILL.md` in WP-F6: forge now mentions "route to founder230first for pre-execution intent" in its clarifying-questions step and its Red Flags.231232---233234## Reference Files235236Read these as needed:237238- `references/fundraising-literacy.md` — pointers to human counsel, questions to ask venture239 lawyers / CFOs, NO actual advice (HR-1)240- `references/legal-questions.md` — trigger list for legal / tax / regulatory questions,241 handoff protocol to human professionals (HR-2)242- `references/venture-brief-schema.md` — complete YAML schema with field semantics, version243 migration notes, validation rules244- `references/forge-handshake.md` — the Scope→Launch handoff contract between founder and forge,245 with the Phase 1 vs Phase 2 split246247---248249## Cross-Cutting Principles250251These apply to ALL founder family interactions regardless of subskill:252253### Grounded, not guru254Every idea, every validation claim, every market observation must cite a real source. If you can't255cite, you can't claim. Default: "I don't know, let me mine some data."256257### Kill criteria or bust258Every idea carries kill criteria + first experiment. An idea without an exit condition is a prayer.259`adversarial-team-brainstorm` enforces this in `founder-ideation`, and the parent refuses to route260forward without them.261262### Adversarial by construction263Generic startup advice is the enemy. The adversarial team primitive exists because LLM-vs-LLM264design produces polished consensus — the one failure mode we're specifically trying to avoid.265Every major output goes through cross-fire.266267### Epistemic honesty as first-class output268"I don't know" and "this needs human review" are valid — and preferred — over false confidence.269HR-10 preserves these in arbiter synthesis.270271### Handoff, don't duplicate272Execution goes to forge. Pitch decks go to presentation-builder. Landing copy goes to273content-writer + conversion-psychology. Founder is a router, not a universal replacement.274275---276277## Anti-Patterns278279| Anti-Pattern | Why It Fails | Correct Approach |280|---|---|---|281| Answering a valuation / cap-table / legal question "just this once" | Creates liability exposure, propagates dangerous misinformation, trains user to rely on LLMs for high-stakes legal decisions | Refuse (HR-1, HR-2). Offer question-prep lists and pointer to human counsel. Never hedge with "I'm not a lawyer but..." — that's the same failure with a fig leaf. |282| Generating TAM numbers from LLM knowledge | All LLM TAM estimates are ungrounded; they look specific but are fabrications | Refuse LLM-generated TAM (HR-3). Calculator mode only — user supplies inputs, skill shows the arithmetic + assumption table. |283| Running `generate_ideas` without MVI captured | Generic-guru output; ideas don't match user's actual context | Enforce MVI at the parent level; refuse to route without biz_type + geography + niche |284| Skipping `adversarial-team-brainstorm` for "simple" ideation | The primitive exists to defeat the polished-guru failure mode; skipping it defeats the founder family's reason for existing | Always use `adversarial-team-brainstorm` via `founder-ideation` for any `generate_ideas` / `evaluate_idea` / `find_niches` request |285| Designing the MVP inside founder | Duplicates forge's job, causes recursion when handoff happens | Stop at Scope→Launch. Hand off to forge. Founder owns PRE-execution, forge owns execution. |286| Ignoring `.founder/venture-brief.yaml` on subskill re-entry | Silently re-derives the company, loses decisions, confuses the user | Every subskill reads on entry, writes on exit (HR-7). Parent verifies the file exists after any subskill call. |287| Returning verbatim Reddit quotes with identifying details | Privacy violation, HR-11 breach, ToS breach | Enforce paraphrase-first at `reddit-signal-mining` level; parent re-filters before displaying to user |288| Hand-waving the forge handshake ("just tell forge to build it") | Loses the structured `forge_brief`, forge has to re-derive context, wastes user time | Populate `forge_brief` fully, set `forge_handoff_ready: true`, invoke forge with the brief as problem statement |289290---291292## When NOT to Use This Skill293294- **Career questions** — "should I leave my job" → `career-transition`, `career-assessment`295- **Implementation / coding** — → `forge` directly (after founder has produced a validated idea)296- **A separate webstore venture** — → `entrepreneur-webstore`297- **Legal / tax / regulatory / securities** — REFUSED; human counsel only298- **Pure writing tasks** (blog posts, landing copy) without founder context → `content-writer`,299 `conversion-psychology`, `seo-*` directly300- **Pitch deck assembly when venture-brief is already validated** — → `presentation-builder`301 directly with the yc-pitch or sequoia-pitch flow (the founder family stubbed these in WP-F8)302- **"How do I build this feature" after execution has started** — → `forge`