
spec
Turn a vague feature or product idea into an agreed, persisted specification through relentless structured questioning. Never assumes — every gap, ambiguity, or "probably" becomes a question to the developer, and the spec cannot be approved while open questions remain. Produces docs/specs/ - .md with acceptance criteria that /plan, /scaffold, and /tdd consume. Use when: "spec", "write a spec", "spec this out", "requirements", "PRD", "acceptance criteria", "define the feature", "user stories", "what should we build", or before planning any feature too big to describe in one sentence.
- Installs
- 2.8K
- Last 24h
- +55
- Trending rank
- #3173
- Security
- Not audited
Install
Works with every agent the skills CLI supports.
npx skills add https://github.com/codewithmukesh/dotnet-claude-kit --skill spec- 01
Install
Run the command in your project. The CLI asks which agents to install for.
- 02
Describe the task
Ask your agent for something this skill covers. It loads spec automatically when relevant.
- 03
Review the result
The agent follows the SKILL.md steps, scripts and references bundled with the skill.
SKILL.md
The instructions your agent reads when the skill is triggered.
/spec — Relentless Specification Workflow
What
Converts an idea into a written, versioned specification that both the developer and Claude explicitly agree on — before any planning or code. The contract:
- Never assume. Every gap in the idea becomes a question. If Claude catches itself thinking "probably", "presumably", or "the usual way" — that thought is a question to ask, not a decision to make.
- Relentless, but structured. Questions come in focused rounds (3–5 at a time) across nine dimensions — not one overwhelming dump, and not a single polite round that stops early.
- Agreement is explicit. Specs have a status lifecycle: Draft → In Review → Approved. Implementation never starts from a Draft. Approval requires the developer to read the final document and say so.
- Specs are files, not chat. Output persists to
docs/specs/<NNN>-<slug>.mdand survives the session. Plans, tests, and commits reference it.
When
- Any feature too big to describe completely in one sentence
- New product or module ideas ("I want to add team workspaces")
- Before
/planfor non-trivial features — plan consumes the approved spec - When requirements feel fuzzy mid-implementation: stop,
/spec, re-plan - Trigger phrases: "spec", "requirements", "PRD", "define the feature", "acceptance criteria"
Skip for: bug fixes, refactors, single-endpoint CRUD where the entity is obvious.
How
Step 1: Capture and Restate
Take the raw idea and restate it in one paragraph: what Claude understood, in its own words. End with: "Is this the idea? What did I get wrong?" Do not begin questioning until the developer confirms the restatement — questioning the wrong idea wastes everyone's time.
Step 2: Questioning Rounds
Work through the nine dimensions in order. Each round: pick the 3–5 most load-bearing unanswered questions (answers that reshape later questions come first). Where the harness supports selectable options, present choices with trade-offs — and a recommendation — but the developer chooses; a recommendation is never silently applied.
| # | Dimension | What to pin down |
|---|---|---|
| 1 | Problem & users | Who hurts today, how they work around it, what success looks like |
| 2 | Scope | What is IN this iteration, what is explicitly OUT, where the MVP line sits |
| 3 | Domain & data | Entities, relationships, lifecycle (create→archive→delete?), retention |
| 4 | API contract | Resources, endpoints, request/response shapes, pagination, versioning |
| 5 | Authorization | Who can do what, role/claim model, tenant boundaries |
| 6 | Edge cases & failure modes | Concurrency, duplicates, idempotency, partial failure, limits |
| 7 | Non-functionals | Expected volume, latency budget, growth assumptions |
| 8 | Integrations | External services, published events, webhooks, side effects |
| 9 | Acceptance criteria | Testable Given/When/Then for every behavior in scope |
Rules of relentless questioning:
- Record every answer in the draft spec immediately — answers are requirements, not conversation.
- Challenge contradictions on the spot: "In round 1 you said X; this answer implies not-X. Which wins?"
- "I don't know" is a legal answer → moves to Deferred Decisions with an explicit fallback the developer chooses now ("default to soft-delete until decided"). Silent deferral is forbidden.
- A dimension is done when a follow-up round generates zero new questions for it.
- The questioning phase is done when ALL nine dimensions are done. Do not stop because the conversation feels long — stopping early is how assumptions sneak in.
Step 3: Draft the Spec File
Determine the next number from existing files in docs/specs/ (create the
directory if missing). Write docs/specs/<NNN>-<slug>.md:
# Spec NNN: <Title>
**Status:** Draft
**Date:** <today>
## Problem <!-- who hurts, why now -->
## Scope <!-- ### In / ### Out — both explicit -->
## Domain Model <!-- entities, relationships, lifecycle -->
## API Contract <!-- endpoints table: verb, route, request, response, errors -->
## Authorization <!-- who can do what -->
## Edge Cases & Failure Modes
## Non-Functional Requirements
## Integrations
## Acceptance Criteria <!-- Given/When/Then, numbered AC-1, AC-2, ... -->
## Deferred Decisions <!-- decision, chosen fallback, revisit trigger -->
## Open Questions <!-- MUST be empty before approval -->
Step 4: Review Loop
Set status to In Review. Present the complete spec and ask: "Read this end-to-end. What is wrong, missing, or over-engineered?" Fold corrections in and re-present. Repeat until the developer has no further changes. New answers may spawn new questions — that is the process working, not a failure to converge.
Step 5: The Agreement Gate
Approval is a deliberate act, never inferred from silence or "looks good" in
passing. Ask explicitly: "Do you approve this spec? After approval, code follows
the spec — changes go through the spec first." On approval, set
**Status:** Approved (<date>).
- Open Questions must be empty. If any remain, the spec cannot be approved — resolve or defer each one explicitly.
- If implementation later reveals a wrong assumption: stop, set status back to In Review, fix the spec with the developer, then resume. Code never silently diverges from an approved spec.
Step 6: Handoff
/planreads the approved spec and maps acceptance criteria to implementation steps/tddturns acceptance criteria into the first failing tests (AC-n → test name)- Commits for the feature reference the spec:
feat: team workspaces (spec 004)
Example
User: /spec users should be able to invite teammates
Claude: Here's what I understood: an existing user can invite another person
to join their... workspace? account? That's already my first gap.
Is this the idea? What did I get wrong?
User: yes, invite to their organization
Claude: Round 1 — Scope & Users:
1. Can invitees be people without an existing account, or only registered users?
2. Who can invite — any member, or only admins/owners?
3. Is there a limit on pending invites (per org, per plan)?
4. Out of scope check: bulk invites and SSO-provisioned members — both OUT for now?
User: 1: both. 2: admins only. 3: no limit. 4: correct, both out.
Claude: Round 2 — Edge cases (your "both" in Q1 raises these):
1. Inviting an email that already has a pending invite — error, or resend?
2. Invite expiry — 7 days, 30 days, never?
3. Invitee already belongs to another org — multi-org membership or rejection?
...
[…rounds continue until all nine dimensions are exhausted…]
Claude: Spec written to docs/specs/004-team-invitations.md (In Review).
Read it end-to-end. What is wrong, missing, or over-engineered?
User: looks right. approved.
Claude: Status: Approved (2026-06-12). Next: /plan to map AC-1..AC-9
to implementation steps.
Related
/plan— Consumes the approved spec; never plan a spec-worthy feature without one/tdd— Acceptance criteria become the first failing tests/scaffold— Generates the slices the plan calls forarchitecture-advisor— Load during Step 2 if the feature forces architectural decisions
FAQ
What does the spec skill do?
Turn a vague feature or product idea into an agreed, persisted specification through relentless structured questioning. Never assumes — every gap, ambiguity, or "probably" becomes a question to the developer, and the spec cannot be approved while open questions remain. Produces docs/specs/ - .md with acceptance criteria that /plan, /scaffold, and /tdd consume. Use when: "spec", "write a spec", "spec this out", "requirements", "PRD", "acceptance criteria", "define the feature", "user stories", "what should we build", or before planning any feature too big to describe in one sentence.
How do I install spec?
Run npx skills add https://github.com/codewithmukesh/dotnet-claude-kit --skill spec in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/spec/ (personal) or .claude/skills/spec/ (project).
Which AI agents can use spec?
Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.
Is spec safe to install?
No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.
Who maintains spec?
It is published by codewithmukesh in codewithmukesh/dotnet-claude-kit. It has 2,831 recorded installs.
More from codewithmukesh/dotnet-claude-kit
clean-architecture
codewithmukesh/dotnet-claude-kit
Clean Architecture for .NET applications. Covers the 4-project layout (Domain, Application, Infrastructure, Api), dependency inversion, use case handlers, domain entities with behavior, and infrastructure as a plugin. Load this skill when building a project with Clean Architecture, discussing layered architecture, dependency inversion, use cases, or when the architecture-advisor recommends Clean Architecture.
ddd
codewithmukesh/dotnet-claude-kit
Domain-Driven Design tactical patterns for .NET applications. Covers aggregates, aggregate roots, value objects, domain events, domain services, strongly-typed IDs, and repository patterns for aggregate persistence. Load this skill when implementing DDD, working with aggregates, value objects, domain events, bounded contexts, or when the architecture-advisor recommends DDD + Clean Architecture. Pair with the clean-architecture skill.
modern-csharp
codewithmukesh/dotnet-claude-kit
Modern C language features for .NET 10 and C 14. Covers primary constructors, collection expressions, the field keyword, extension members, records, pattern matching, spans, and raw string literals. Load this skill when writing any new C code, reviewing existing code for modernization, using "modern C", "C 14", "primary constructor", "collection expression", "records", "pattern matching", "span", "field keyword", or "extension members". Always loaded as the baseline for all agents.
vertical-slice
codewithmukesh/dotnet-claude-kit
Vertical Slice Architecture (VSA) for .NET applications — one of several supported architectures in dotnet-claude-kit. Covers feature folders, endpoint grouping, and handler patterns for Mediator, Wolverine, and raw handler classes. Load this skill when the architecture-advisor recommends VSA, when working in an existing VSA codebase, when adding features to a feature-folder project, or when discussing vertical slice patterns, feature folders, or handler patterns.
security-scan
codewithmukesh/dotnet-claude-kit
Deep security scanning for .NET applications across 6 layers: vulnerable packages, secrets detection, OWASP code patterns, auth configuration, CORS policy, and data protection. Produces severity-rated findings with specific remediation steps. Load this skill when: "security scan", "security audit", "check for vulnerabilities", "find secrets", "OWASP", "auth review", "CORS check", "security review", "penetration test prep", "CVE check", "vulnerability scan", "hardcoded password", "data protection", "security posture".
serilog
codewithmukesh/dotnet-claude-kit
Structured logging with Serilog for .NET 10 applications. Covers two-stage bootstrap, appsettings configuration, enrichers, sinks, request logging, destructuring, and Serilog.Expressions. Load this skill when setting up Serilog, configuring log sinks, enrichers, or structured logging, or when the user mentions "Serilog", "structured logging", "log enrichment", "Seq", "LogContext", "UseSerilog", "WriteTo", "message template", "Serilog.Expressions", "request logging", "log sink", "rolling file", or "audit log".
Popular code quality & workflow skills
grill-me
mattpocock/skills
A relentless interview to sharpen a plan or design.
improve-codebase-architecture
mattpocock/skills
Scan a codebase for deepening opportunities, present them as a visual HTML report, then grill through whichever one you pick.
handoff
mattpocock/skills
Compact the current conversation into a handoff document for another agent to pick up.
domain-modeling
mattpocock/skills
Build and sharpen a project's domain model. Use when discussing codebase terminology, writing or editing a GLOSSARY.md, or recording or editing an ADR.
codebase-design
mattpocock/skills
Shared vocabulary for designing deep modules. Use when the user wants to design or improve a module's interface, find deepening opportunities, decide where a seam goes, make code more testable or AI-navigable, or when another skill needs the deep-module vocabulary.
diagnosing-bugs
mattpocock/skills
Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.