메뉴
Smart pre-build discovery that validates novel feature ideas before committing to a full PRD. Classifies objectives as common patterns (skip) or novel concepts (full validation with lean canvas, competitive scan, and risk assessment).
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
Interactive setup wizard for Company OS. Configures company.config.yaml, generates .claude/settings.json with tech-stack-specific permissions, and scaffolds directories. Supports three modes — interactive wizard, express config block, or auto-extract from a URL/unstructured text.
Produces RFC/ADR defining system boundaries, modules, data model, and integration points. Use when translating a PRD into technical design, making architectural decisions, or designing new system components.
Produces architecture diagrams, data flow diagrams, ERDs, and system overviews in Mermaid (inline) and Excalidraw JSON (standalone). Use when an RFC, threat model, or documentation needs visual diagrams.
Performs data flow threat analysis with structured mitigations. Use when assessing security threats for new features, architecture changes, or compliance reviews.
Creates platform-specific marketing playbooks for acquisition channels. Use when launching or optimizing a specific growth channel (e.g., LinkedIn, Twitter, email, Product Hunt).
Prepares pre-call briefings, captures post-call debriefs, and routes insights to ICP refinement and PRD evidence. Stores conversation logs as CONV- artifacts in artifacts/decision-memos/.
SOC 직업 분류 기준
| name | discovery-validation |
| description | Smart pre-build discovery that validates novel feature ideas before committing to a full PRD. Classifies objectives as common patterns (skip) or novel concepts (full validation with lean canvas, competitive scan, and risk assessment). |
| user-invokable | false |
Not every feature needs discovery. Authentication, CRUD, and notifications are well-understood patterns -- just build them. But novel concepts (new business models, AI-powered interfaces, marketplace dynamics) carry real uncertainty and deserve lightweight validation before committing engineering time. This skill acts as a smart filter: it classifies the objective, skips discovery for common patterns, and runs a focused validation for novel concepts.
Determine whether this feature needs discovery by matching it against known patterns.
These are well-understood, solved problems with established implementation patterns:
If the objective matches a common pattern:
Output: "Discovery skipped -- [feature] matches common pattern: [category]. Proceeding to PRD."
Exit the skill. No further steps needed.
These involve genuine uncertainty about problem-solution fit:
If the objective matches a novel concept, proceed to Step 2.
If the objective does not clearly fit either category, ask the user:
"This feature sits between common and novel. Should I run discovery validation or proceed directly to PRD?"
Respect the user's decision.
Produce a one-page lean canvas covering all nine boxes. Be specific -- generic answers defeat the purpose.
| Box | Prompt | Output |
|---|---|---|
| Problem | What are the top 3 problems this feature solves? | Specific, observable problems with evidence |
| Customer Segments | Who specifically has these problems? | Named persona(s) with context (role, company size, behavior) |
| Unique Value Proposition | Single clear compelling message | One sentence that would make the target persona stop scrolling |
| Solution | Top 3 features that address the problems | Concrete capabilities, not abstract descriptions |
| Channels | How will users discover this feature? | Specific channels (in-app prompt, email campaign, blog post, word-of-mouth) |
| Revenue Streams | How does this feature contribute to revenue? | Direct (new pricing tier, add-on) or indirect (retention, upsell trigger) |
| Cost Structure | What are the build + ongoing costs? | Engineering time estimate, infrastructure costs, third-party service costs |
| Key Metrics | What numbers prove this works? | 2-4 specific metrics with target values |
| Unfair Advantage | What makes this defensible? | Data, network effects, domain expertise, integrations, brand |
Identify 3-5 closest alternatives and analyze them:
For each alternative:
Gap Analysis: What is missing from existing solutions? What do users complain about? Where is there whitespace?
Positioning: How will our approach be different or better? What trade-offs are we making? Why would someone switch from an existing solution?
Use WebSearch to find current competitors, recent reviews, and user complaints if needed. Do not fabricate competitor information -- if data is unavailable, state that explicitly.
Identify the ONE thing that, if wrong, kills this idea. Be specific:
How to test the riskiest assumption cheaply, before building:
| Question | How to Answer |
|---|---|
| Who has this problem? | Specific persona with demographics and behavior |
| How painful is it? (1-10 scale) | Evidence-based: time wasted, money lost, frustration signals |
| How are they solving it today? | Current workarounds, tools, manual processes |
| Will they pay for a better solution? | Willingness-to-pay signals: existing spend, stated intent, market comps |
Based on the lean canvas, competitive scan, and risk assessment, provide one of three recommendations:
Include a clear rationale for the recommendation (2-3 sentences).
Append a "Discovery Validation" section to the PRD (this is NOT a separate artifact -- it lives within the PRD):
## Discovery Validation
**Classification**: NOVEL -- [reason why this needed discovery]
**Lean Canvas**: [summary of the 9 boxes, or "See full canvas above"]
**Competitive Landscape**: [3-5 alternatives with gap analysis summary]
**Riskiest Assumption**: [the one specific thing that could kill this]
**Validation Status**: Validated / Needs Testing / Risky
**Recommendation**: Proceed to PRD / Validate first / Reconsider
**Rationale**: [2-3 sentences explaining the recommendation]
If discovery was skipped (common pattern), the PRD should include a brief note:
## Discovery Validation
**Classification**: COMMON PATTERN -- [category name]
**Discovery**: Skipped. This is a well-understood pattern with established implementation approaches.