speccy（speccy 搜索）

Speccy - Interview-Driven Specification Builder

When this skill is invoked, IMMEDIATELY output the banner below before doing anything else. Pick ONE tagline at random — vary your choice each time. CRITICAL: Reproduce the banner EXACTLY character-for-character. The first line of the art has 4 leading spaces — you MUST preserve them.

{tagline}

⠀   ██╗███████╗██████╗ ███████╗ ██████╗ ██████╗██╗   ██╗
   ██╔╝██╔════╝██╔══██╗██╔════╝██╔════╝██╔════╝╚██╗ ██╔╝
  ██╔╝ ███████╗██████╔╝█████╗  ██║     ██║      ╚████╔╝
 ██╔╝  ╚════██║██╔═══╝ ██╔══╝  ██║     ██║       ╚██╔╝
██╔╝   ███████║██║     ███████╗╚██████╗╚██████╗   ██║
╚═╝    ╚══════╝╚═╝     ╚══════╝ ╚═════╝ ╚═════╝   ╚═╝

Taglines:

• 🔍 Tell me everything...
• 🧠 Let's think this through!
• 📋 Spec it before you wreck it!
• 🎤 Interview mode: ACTIVATED
• 💡 Great specs start with great questions!
• 🏗️ Measure twice, code once!
• 📝 No assumption left behind!
• 🎯 Precision engineering starts here!

Output Formatting

After the banner, display parsed input:

┌─ Input ────────────────────────────────────────
│  {Field}:  {value}
│  Flags:    {parsed flags or "none"}
└────────────────────────────────────────────────

Pre-flight results:

── Pre-flight ───────────────────────────────────
  ✅ {dep}           {version or "found"}
  ⚠️ {dep}           not found → {fallback detail}
  ❌ {dep}           missing → stopping
──────────────────────────────────────────────────

Stage/phase headers: ━━ {N} · {Name} ━━━━━━━━━━━━━━━━━━━━━━━━━

Status icons: ✅ done · ❌ failed · ⚠️ degraded · ⏳ working · ⏭️ skipped

Interview the user through multiple rounds of targeted questions to build a comprehensive specification, then write it directly using the spec template in references/spec-template.md.

Interview prompts and question guidelines: references/interview-guide.md Spec template and writing guidelines: references/spec-template.md

Pre-flight

Before starting, check all dependencies in this table:

For each row, in order:

1. Test file existence (check both paths for symlinked skills)
2. If found: continue silently
3. If missing: apply Resolution strategy
4. After all checks: proceed to context gathering

Stage 1: Context Gathering

Before asking any questions, build a thorough understanding of the project:

1. Capture GOAL — the user's argument describing what needs to be specified
2. Load project context — invoke /prime to load domain-specific context (CLAUDE.md, specs, memory). If /prime is unavailable, fall back to the manual scan below.
3. Scan the project (skip items already loaded by /prime):

- Read CLAUDE.md if present (project conventions, structure, domain) - Scan specs/ directory for existing specifications - Scan existing design docs for context - Read relevant source code that relates to the GOAL - Check memory for prior decisions or open questions related to the GOAL

1. Identify knowledge gaps — what must you learn from the user to write a complete, unambiguous specification?

Group gaps into interview categories:

• Architecture & Technical Design — stack, patterns, data flow, integrations
• Requirements & Scope — what's in, what's out, must-haves vs nice-to-haves
• UI & UX — user flows, interaction patterns, accessibility, responsive
• Security & Auth — authentication, authorization, data protection
• Infrastructure & Deployment — hosting, CI/CD, environments, IaC
• Data & Storage — schemas, persistence, migrations, caching
• Testing & Quality — test strategy, coverage, acceptance criteria
• Concerns & Tradeoffs — known risks, alternatives considered, constraints

Stage 2: Interview Rounds

Conduct multiple rounds of questions using AskUserQuestion. Continue until all knowledge gaps are resolved.

Question Rules

1. 4 questions per round maximum (AskUserQuestion limit)
2. Non-obvious questions only — don't ask things you can determine from reading the code or docs. The user's time is valuable.
3. Recommendations — where you have an informed opinion based on the codebase, project conventions, or industry best practice, mark one option as recommended by listing it first and appending (Recommended) to its label. At least one question per round should have a recommendation where possible.
4. Concise options — 2-4 options per question, each with a clear description of implications and tradeoffs
5. Progressive depth — start with high-level architecture and scope, then drill into implementation details in later rounds
6. Build on answers — use previous round answers to inform next questions. Don't re-ask decided topics.
7. Track decisions — maintain a running list of all decisions made. Present this list at the start of each round so the user can see progress.

Round Structure

Each round follows this pattern:

1. Progress update — brief summary of decisions made so far (after round 1)
2. Category label — which interview category this round covers
3. Questions — 3-4 targeted questions via AskUserQuestion
4. Evaluate — after answers, determine if more questions are needed

Completion Criteria

Stop interviewing when ALL of the following are true:

• All identified knowledge gaps have been addressed
• No answer has raised new unresolved questions
• You have enough information to write every section of the spec template
• The user has confirmed scope boundaries (what's in and what's out)

When complete, briefly present a Decision Summary — a numbered list of all decisions made across all rounds — and confirm with the user before proceeding to spec generation.

Stage 3: Generate Specification

Once the interview is complete and decisions are confirmed:

1. Create specs/ directory if it doesn't exist: mkdir -p specs
2. Read the spec template from references/spec-template.md
3. Generate the spec by filling the template with:

- The original GOAL as the introduction and purpose - All decisions from the interview rounds, mapped to the appropriate sections - Code/architecture context discovered in Stage 1 - Acceptance criteria derived from requirements decisions - Test strategy aligned with the project's existing patterns

1. Write the spec file to specs/{name}.md where {name} is a kebab-case slug derived from the GOAL (e.g., specs/user-auth.md, specs/payment-integration.md). Use the Write tool directly.
2. The specs/ directory is the standard location — /build and /prime both scan it automatically.

Output & Handoff

After the spec is created, report to the user:

┌─ Speccy · Report ──────────────────────────────
│
│  ✅ Spec complete
│
│  📄 File:       {spec file path}
│  📋 Sections:   {count}
│  💬 Rounds:     {interview rounds conducted}
│  ❓ Questions:  {total questions asked}
│
│  📝 Key decisions
│     • {decision 1}
│     • {decision 2}
│     • {decision 3}
│
│  🔗 Links
│     Spec: {spec file path}
│
│  ⚡ Next steps
│     1. Review the spec: {path}
│     2. Run `/build {spec path}` to implement (reads the file automatically)
│
└─────────────────────────────────────────────────

Then immediately ask the user via AskUserQuestion:

Question: "Spec written to {spec file path}. Ready to build?" Options:

• "Clear context & build (Recommended)" — clear the conversation first, then run /build
• "Build now" — invoke /build immediately in the current context
• "Review first" — stop here so the user can review the spec before building
• "Done" — stop here, no build

If the user selects "Clear context & build":

1. Run /clear to clear the conversation context
2. Then invoke /build {spec file path}: Skill(skill: "build", args: "{spec file path}") /build reads the spec file via Plan Resolution and executes the full pipeline with maximum context window available.

If the user selects "Build now":

1. Invoke /build {spec file path} directly: Skill(skill: "build", args: "{spec file path}") /build reads the spec file via Plan Resolution and executes the full pipeline. The current conversation context is preserved.

In both cases, do not attempt to implement the spec yourself — always delegate to /build.

If the user selects "Review first" or "Done", stop and display:

⚡ To build later, run: /build {spec file path}

IMPORTANT: After generating the spec, do NOT enter plan mode, do NOT start implementing directly, and do NOT offer to execute the plan yourself. The only path to implementation is through /build.