Feature Spec

Creates structured feature specifications and PRDs from problems, ideas, and requirements.

Feature Spec is an official product-management skill from Anthropic’s knowledge-work-plugins. The original source was expected at product-management/skills/feature-spec. In the currently reachable main branch, the workflow is published at product-management/skills/write-spec; the candidate name feature-spec still describes the same documented purpose. The direct primary source for the current content is https://github.com/anthropics/knowledge-work-plugins/tree/main/product-management/skills/write-spec. According to the provider, the skill turns a feature, problem statement, user request, or still-vague idea into a structured feature specification or Product Requirements Document.

Problem and context

The workflow starts with a clear statement of what needs to be specified. It then gathers the user problem, affected target segments, desired success criteria, technical or organizational constraints, and prior attempts or solutions. This conversation is intentionally not a single large questionnaire. According to the provider, the most important uncertainties should be addressed first, with missing details filled in as the discussion develops. When connected project, knowledge, design, or feedback sources are available, the workflow can incorporate related tickets, research, mockups, meeting notes, and user feedback. Without those connections, it works entirely from the context supplied by the team.

Output structure

The recommended specification contains a problem statement describing affected people and the cost of inaction, measurable goals, and explicit non-goals. User stories describe specific roles, capabilities, and benefits without prescribing a particular implementation. Requirements are grouped into must-have elements, useful additions, and future considerations. Important requirements receive testable acceptance criteria, often expressed in Given-When-Then form. Success measurement distinguishes leading signals such as adoption, activation, completion rate, time to complete, usage frequency, and errors from lagging signals such as retention, satisfaction, support reduction, competitive outcomes, or business impact. Open questions are assigned to engineering, design, legal, data, or stakeholder owners. Timeline considerations capture hard deadlines, dependencies, and phasing.

Prioritization and scope

A major benefit is deliberate scope control. Non-goals prevent gradual expansion, while P0, P1, and P2 categories make clear what belongs in the first viable release, what can follow, and what should influence future architecture. The source also presents MoSCoW as a possible prioritization framework. A useful document records assumptions, risks, and dependencies instead of presenting uncertain decisions as facts. Acceptance criteria should cover happy paths, error states, empty states, and boundary conditions. Vague words such as fast or intuitive should be replaced by observable conditions. The specification is a shared basis for alignment and delivery, not an immutable contract.

Boundaries, safety, and fit

Feature Spec is not project-management software, product analytics, an autonomous prioritization decision, or an approval system. According to the provider, the generated document should be reviewed and iterated with stakeholders after it is produced. A model response cannot replace user research, reliable measurement, legal review, or engineering investigation. Connected sources must be used only within approved permissions. Internal tickets, personal feedback, and confidential product plans should be minimized and handled according to the applicable access policy. A local skill directory does not mean that prompts and results remain local; a connected model may transmit data to its model provider. Credentials, tokens, and production data do not belong in examples. According to the provider, Anthropic publishes the repository under Apache-2.0. This description was reviewed on September 9, 2026 against the official repository, its product-management overview, and official Claude Code documentation. It describes the documented workflow, not a guarantee of correct requirements or successful product decisions.

Installation instructions

What Feature Spec is and why it matters

Behind the catalog entry Feature Spec sits the write-spec skill from the open-source product-management plugin, part of Anthropic's knowledge-work-plugins collection. The skill helps turn a vague idea, a problem statement, or a user request into a structured feature concept or product requirements document, commonly called a PRD. A PRD is a document widely used in product teams that describes which problem a planned feature solves, what goals it should achieve, what requirements exist, and how success will be measured. Per the provider, the entire product-management plugin is aimed at roles like product managers, helping them write specs, plan roadmaps, synthesize user research, and keep stakeholders updated. The skill is closely tied to the parent product Claude Cowork as well as to Claude Code, since both platforms can use the same plugins, and the skill is invoked there as the slash command write-spec.

Step 1: Prerequisites

Using the skill requires either Claude Cowork or Claude Code with access to the plugin marketplace. No deep technical background is strictly necessary, since the skill actively asks the user for the relevant information. Optionally, connectors to project management tools such as Linear, Asana, or Jira, as well as knowledge bases like Notion or design tools like Figma, can be configured so the skill automatically pulls in existing tickets, research, or drafts. These connectors are explicitly optional, though; per the documentation, the skill also works entirely without any connected tools, relying solely on what the user provides during the conversation.

Setting it up step by step

First, the plugin marketplace is added with the command claude plugin marketplace add anthropics knowledge-work-plugins, after which the product-management plugin is installed specifically via claude plugin install product-management at knowledge-work-plugins. Once the plugin is active, the skill is invoked through the slash command write-spec, either by directly supplying a feature description or by letting the skill ask questions step by step. The skill asks in sequence about the underlying user problem, the affected target audience, possible success metrics, technical or scheduling constraints, and any prior attempts at solving the same problem, before generating a complete document containing a problem statement, goals, explicit non-goals, user stories, prioritized requirements, success metrics, and open questions.

Security and best practices

Whenever connectors to external tools such as Slack, Notion, or a ticketing system are enabled, the skill can search those tools for relevant context, so only the access rights actually needed should be granted. Before sharing the generated document externally, it is worth reviewing the automatically pulled-in context for confidential information, such as internal metrics or customer data that may have been carried over unintentionally from connected systems. Since the skill only produces text and does not execute code, the technical risk is low; still, the generated document does not replace a substantive review by a product team, especially where success metrics and effort estimates are merely suggested by the AI.

Practical example and limitations

For example, entering the user request that customers want to export their data as a CSV file produces a structured document with prioritized requirements, clearly separated non-goals, and concrete user stories for each audience segment. The skill's limitation is that it does not itself create tickets in a project management system; it merely offers to generate further artifacts afterward, such as a design brief or an engineering task breakdown. The resulting PRD should therefore be understood as a solid first draft that still needs to be reviewed, prioritized, and finalized by a human product team.

Free
Provider
Anthropic
License
Apache-2.0
Last reviewed
09.09.2026

Repository and documentation

Categories

Compatible with

Claude Code