| name | business-idea-validation |
| description | Sequential stage-gate pipeline for deep business idea validation. 9 stages, 6 specialist sub-agents, kill gates between every stage. Takes an idea from raw concept through problem validation, customer analysis, market sizing, competitive intel, customer simulation, synthesis, GTM strategy, financial viability, to a scored Go/No-Go decision. Each stage builds on all prior artifacts — the pipeline is strictly sequential, not parallel.
|
| tags | ["business","validation","startup","pipeline","sub-agents","orchestration","stage-gate"] |
| alwaysApply | false |
Business Idea Validation — sequential stage-gate pipeline
You are orchestrating a validation pipeline: a sequence of specialist
sub-agents that each execute one stage of business idea validation, producing
a cumulative dossier. Unlike multi-review (parallel fan-out) or council
(parallel rounds), this pipeline is strictly sequential — each stage
depends on all prior artifacts. Between every stage, you evaluate a kill gate
that can halt the pipeline early.
When to invoke
Trigger phrases / explicit invocations:
/validate-idea- "validate this business idea"
- "run idea validation"
- "is this idea worth pursuing?"
- "go/no-go on this concept"
- "business idea validation"
- "validate my startup idea"
- "stage-gate this idea"
Use it for: deep validation of a specific business idea — the user already
has a concept and wants to stress-test it before committing time/money.
Do NOT use it for:
- Finding ideas from scratch → use
business-idea-discovery skill.
- Quick vibes check → just answer directly.
- Post-build growth → use
growth skill.
- Technical feasibility review → use
engineering skill.
Relationship to business-idea-discovery
business-idea-discovery = find ideas fast (single agent, 2-5 hours).
business-idea-validation (this skill) = validate one idea deeply (6 specialist
agents, sequential pipeline, kill gates). They're complementary:
discover first, then validate the best candidate.
Flow (follow this exactly)
0. Capture the brief
Ask the user:
- What's the idea? (product/service description)
- Who is it for? (target audience, if known)
- What problem does it solve? (core pain, if known)
- Ambition level? — VC-scale ($100M+) or bootstrapped profitable ($1-10M)?
This affects market sizing gates.
- Do you have any existing research? (interviews, data, prior analysis)
Write the answers to _biv/{idea-slug}/00-brief.md. The slug is kebab-case
derived from the idea name (e.g., ai-loyalty-programs).
Initialize _biv/{idea-slug}/pipeline-state.json:
{
"idea": "<slug>",
"ambition": "vc | bootstrapped",
"current_stage": 0,
"stages": {
"1": { "status": "pending", "confidence": null, "gate": null },
"2": { "status": "pending", "confidence": null, "gate": null },
"3": { "status": "pending", "confidence": null, "gate": null },
"4": { "status": "pending", "confidence": null, "gate": null },
"5": { "status": "pending", "confidence": null, "gate": null },
"6": { "status": "pending", "confidence": null, "gate": null },
"7": { "status": "pending", "confidence": null, "gate": null },
"8": { "status": "pending", "confidence": null, "gate": null },
"9": { "status": "pending", "confidence": null, "gate": null }
},
"started_at": "<ISO timestamp>",
"completed_at": null,
"verdict": null
}
1-9. Execute stages sequentially
For each stage (1 through 9), follow this protocol:
a. Dispatch the specialist
Dispatch the appropriate sub-agent (see Agent-Stage Mapping below) in
foreground (not background — you need the result before proceeding).
The sub-agent's prompt MUST include:
- The idea brief (
00-brief.md contents)
- All prior stage artifacts (read each
{NN}-*.md file and include contents)
- The stage number and what it must produce
- The ambition level (affects scoring thresholds)
- Any user feedback from a prior WEAK gate iteration
b. Read the sub-agent's output
The sub-agent writes its artifact to _biv/{idea-slug}/{NN}-{stage-name}.md
and returns an executive summary + confidence score + any kill signals.
c. Evaluate the gate
Read the sub-agent's structured output. Apply the gate:
| Gate Result | Condition | Action |
|---|
| PASS | Confidence ≥ 3, no kill signals | Update pipeline-state.json, show exec summary to user, proceed |
| WEAK | Confidence = 2, no kill signals | Show findings to user. Ask: "Iterate this stage with adjustments, or proceed with acknowledged risk?" |
| KILL | Confidence ≤ 1 OR kill signals present | Stop pipeline. Show findings + kill reasons. Recommend: pivot the weak dimension or abandon the idea. |
If WEAK and user says "iterate": re-dispatch the same sub-agent with the
user's feedback appended. Max 2 iterations per stage — after that, force
a decision (proceed or kill).
If WEAK and user says "proceed": mark gate as "passed_with_risk" and continue.
d. Update state
After each gate decision, update pipeline-state.json with:
status: "completed" / "passed_with_risk" / "killed"confidence: the sub-agent's score (1-5)gate: "pass" / "weak_proceed" / "weak_iterate" / "kill"
Agent-Stage Mapping
| Stage | # | Agent | Model |
|---|
| Problem Discovery | 1 | biv-researcher | sonnet |
| Customer Analysis | 2 | biv-customer-analyst | sonnet |
| Market Sizing | 3 | biv-market-analyst | sonnet |
| Competitive Intelligence | 4 | biv-researcher | sonnet |
| Customer Validation (simulated) | 5 | biv-customer-analyst | sonnet |
| Synthesis & Hypothesis | 6 | biv-strategist | opus |
| GTM Strategy | 7 | biv-strategist | sonnet |
| Financial Viability | 8 | biv-financial-analyst | sonnet |
| Go/No-Go Decision | 9 | biv-decision-gate | opus |
Note: biv-researcher is called at stages 1 AND 4 with different prompts.
biv-customer-analyst is called at stages 2 AND 5 with different prompts.
biv-strategist is called at stages 6 AND 7 with different prompts.
biv-strategist defaults to opus — pass model: sonnet explicitly when
dispatching for Stage 7 (GTM is less synthesis-heavy than Stage 6).
10. Present the verdict
After Stage 9 completes, read 09-go-no-go-decision.md and present:
- The scorecard (8 dimensions, each scored 1-5)
- The total score and verdict: GO (30-40) / PIVOT (18-29) / NO-GO (8-17)
- A pipeline summary: which stages passed clean, which had risk, which were strong
If GO: offer to generate a Lean Canvas as a capstone (re-dispatch biv-strategist
to produce 10-lean-canvas.md from all artifacts).
If PIVOT: identify the weakest dimension(s) and recommend which stage to re-run.
If NO-GO: frame it as a success — "you just saved months of wasted effort."
Resume protocol
If a conversation breaks mid-pipeline:
- Read
_biv/{idea-slug}/pipeline-state.json
- Identify
current_stage and which stages are completed
- Tell the user: "Pipeline for '{idea}' is at Stage {N}. Stages 1-{N-1} are
complete. Resuming from Stage {N}."
- Continue from the next pending stage
The user can also explicitly say "re-run stage 4" to re-execute a specific stage
with updated context.
Hard rules
- Stages run strictly sequentially. Never dispatch two stages in parallel.
Every stage depends on prior artifacts.
- You evaluate gates, not the sub-agents. Sub-agents produce artifacts and
confidence scores. You decide PASS/WEAK/KILL based on the scoring rubric.
- All artifacts go to disk. Every stage writes
_biv/{slug}/{NN}-name.md.
This is the cumulative dossier — it persists across conversations.
- Sub-agent prompts include ALL prior artifacts. Stage 6 receives outputs
from stages 1-5. Stage 9 receives outputs from stages 1-8. No stage operates
in isolation.
- Max 2 iterations per stage. A WEAK gate allows one re-run. After 2
attempts, force a proceed-with-risk or kill.
- Context budget: keep your main-agent context under ~20K tokens. You read
executive summaries from sub-agents, not their full research. The full
artifacts are on disk for subsequent sub-agents to consume.
- Stage 5 uses AI simulation. Customer validation is synthesized from
Reddit/forum data collected in Stage 1. The artifact is marked as
"synthetic validation" with an automatic confidence cap of 3 (never higher).
Tell the user that real interviews would increase confidence.
Slash command
/validate-idea invokes this skill. Natural-language triggers listed above
also activate it.
