plan-feature

Shape feature requests into ratified plans with routing, scope, and settled key decisions.

1|Updated Jul 9, 2026
One-click install
npx skills add https://github.com/matthewalton/speccle --skill plan-feature-matthewalton
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: plan-feature
Source: https://github.com/matthewalton/speccle/tree/main/packages/plugin/skills/plan-feature
Command: npx skills add https://github.com/matthewalton/speccle --skill plan-feature-matthewalton

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Feature requests arrive as prose, tickets, or conversations with open decisions and no clear home in the codebase. This Skill turns that raw input into a ratified plan — routing the work to a new slice, an amendment, or a carve — so the rest of the pipeline can execute unattended. ## Core Features & Use Cases - Work routing: Explores existing feature folders and SPEC.md files to decide whether the request is a new slice, an amendment to an owning slice, or a carve of ungoverned behaviour. - Decision settlement: Asks key decisions one at a time with recommendations, captures each into ADRs and CONTEXT.md glossaries as it lands, and flags defaulted decisions in unattended runs. - Plan lenses and summary: Fans out repo-defined plan lenses from .speccle/lenses/plan/ as advisory subagents, then ends with a one-screen plan summary covering route, folder, key, scope, and decisions. - Use Case: A product manager pastes a ticket describing a new discount behaviour; the Skill finds the owning checkout slice, settles the rounding policy with the user, writes the ADR, and hands a readable plan to spec-feature. ## Quick Start Ask the agent to plan a new feature from your ticket or rough description, for example: plan where a discount-code behaviour should live and settle its open decisions before we build it.

Frequently Asked Questions about plan-feature

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I plan a feature before writing code with an AI agent?▼

Provide the request in any form — prose, a ticket, or a conversation — and the Skill explores the repo, routes the work to a new or existing feature slice, settles open key decisions with you one at a time, and ends with a readable plan summary for approval.

How does the skill decide between a new feature and an amendment?▼

Routing is decided by where the behaviour lives: it reads every existing SPEC.md and its criteria, then routes to amend when a governed slice already owns the behaviour, new when none does, and carve when the behaviour runs but is ungoverned.

What is a key decision in feature planning?▼

A key decision is a choice the input leaves open with more than one viable answer that shapes the slice across criteria, such as a rounding policy or data shape. Each is asked one at a time with a recommendation, then captured into an ADR or CONTEXT.md glossary entry.

Can feature planning run unattended without a human?▼

Yes. When no one can answer, the Skill takes its own recommendation and flags the decision as defaulted in the plan summary and at the end of the pipeline, so a defaulted decision is never silent.

What are plan lenses in the .speccle directory?▼

Plan lenses are optional repo-defined markdown prompts in .speccle/lenses/plan/ that review a plan's shape. Each runs as a parallel subagent, and findings are advisory only — they cannot veto the plan and are surfaced in the summary for the human to rule on.

Does the planning skill write code or acceptance criteria?▼

No. It drafts no criteria and writes no code; its only files are decision records such as ADRs and glossary entries. Criteria drafting belongs to the downstream spec-feature stage in the same session.