MCPInstall

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
  1. 01

    Install

    Run the command in your project. The CLI asks which agents to install for.

  2. 02

    Describe the task

    Ask your agent for something this skill covers. It loads spec automatically when relevant.

  3. 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>.md and 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 /plan for 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.

#DimensionWhat to pin down
1Problem & usersWho hurts today, how they work around it, what success looks like
2ScopeWhat is IN this iteration, what is explicitly OUT, where the MVP line sits
3Domain & dataEntities, relationships, lifecycle (create→archive→delete?), retention
4API contractResources, endpoints, request/response shapes, pagination, versioning
5AuthorizationWho can do what, role/claim model, tenant boundaries
6Edge cases & failure modesConcurrency, duplicates, idempotency, partial failure, limits
7Non-functionalsExpected volume, latency budget, growth assumptions
8IntegrationsExternal services, published events, webhooks, side effects
9Acceptance criteriaTestable 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

  • /plan reads the approved spec and maps acceptance criteria to implementation steps
  • /tdd turns 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 for
  • architecture-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

View all →

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.

Backend & APIs2.9K installs

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.

Backend & APIs2.9K installs

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".

Security2.9K installs

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".

DevOps & Cloud2.9K installs

Popular code quality & workflow skills

View category →

grill-me

mattpocock/skills

A relentless interview to sharpen a plan or design.

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.