eho/agent-skills

design-doc

Create or update a concise, repository-grounded design document with canonical user stories ready for independent review and feature delivery.

First seen Mar 24, 2026

Installation

$ npx skills add eho/agent-skills --skill design-doc

Summary

  • Create or update a concise, repository-grounded design document with canonical user stories ready for independent review and feature delivery.
  • Use when asked to write a design doc, PRD, implementation plan, feature spec, requirements, or to turn a technical discussion into an implementable design.

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

Also in this package

Other skills from eho/agent-skills · top by installs.

npx skills add eho/agent-skills

Browse all from eho/agent-skills

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Not declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Repository health

Default branch main
Open issues 0
Status Active

Skill metadata

Parsed from SKILL.md frontmatter.

Version3.0.0
More metadata
author
eho
version
3.0.0

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 5,329 B
  • docs SUMMARY.md 316 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 39 installs

SKILL.md

Design Doc Writer

Turn product and technical context into the smallest design that a capable implementation agent can execute safely.

Read [the shared design contract](references/design-contract.md) completely before writing. It defines readiness, evidence, authority, story syntax, and the downstream feature-delivery contract.

Workflow

  1. Mine the conversation. Extract the problem, goals, constraints,

non-goals, decisions and rationale, rejected directions, and unresolved decisions. Preserve explicit user choices.

  1. Ground the design. Read repository instructions and relevant vision,

architecture, code, tests, schemas, and configuration. Use search before guessing. Verify important paths, current signatures, ownership, and integration points.

  1. Clarify only material gaps. Ask a concise, batched question only when an

answer would materially change scope, architecture, compatibility, safety, or story boundaries and cannot be discovered or safely inferred. Otherwise state the assumption.

  1. Synthesize the design. Explain the chosen behavior and contracts, not

the discovery transcript. Prefer an existing repository pattern unless the design gives a reason to depart from it.

  1. Write canonical stories. Use stable IDs and make every story independently

implementable from its copied issue body. Minimize story count while keeping each story to one coherent, dependency-ordered outcome.

  1. Validate. Resolve design-to-issues/scripts/story_contract.py from the

installed or repository skill set and run:

``bash python3 <story-contract-script> <design-doc> \ --repo-root <repository-root> --mode author ``

Fix every validation failure before reporting completion.

  1. Save as Draft. Write docs/design/<short-kebab-case-name>.md unless the

user or repository specifies another path. New and revised content remains Draft; this skill never promotes it to Revised.

Adaptive document shape

Use only sections that carry useful design information. A typical document is:

# Feature name

**Status:** Draft

## Problem and Goals
## Scope and Non-Goals
## Decisions and Rationale
## Architecture and Contracts
## Integration and Behavior
## Risks and Verification
## Open Questions
## User Stories
## Changelog

Adapt or combine headings to fit the feature. Include security, privacy, data ownership, migration, compatibility, performance, observability, rollout, and rollback where material. Include alternatives when they explain a consequential decision. Omit irrelevant sections instead of filling them with boilerplate.

Use exact examples, data shapes, sequences, or diagrams only when they remove implementation ambiguity. Mark proposed interfaces as proposed; do not present them as current repository facts.

Story guidance

Follow the exact story shape and dependency syntax in the shared contract. Additionally:

  • Put behavioral results in Acceptance Criteria and implementation guidance in

Implementation Context.

  • Make criteria binary and observable. Do not use file placement, lint success,

or vague statements such as “works correctly” as behavioral acceptance criteria.

  • In Verification, name the tests, commands, browser/device checks, migrations,

or operational evidence appropriate to the story. Do not require every verification mode for every story.

  • Include documentation or operational work only when the story changes those

surfaces.

  • When revising shared architecture or contract prose, update every affected

story. Story content is what issue synchronization hashes and delivers.

  • Preserve existing IDs across revisions. Adding, removing, deferring, or

materially rescoping stories must respect the authority rules in the shared contract.

Revision behavior

When asked to address a review artifact, evaluate each blocking finding against the evidence and original intent. Satisfy the required invariant with the smallest coherent change; a reviewer suggestion is not automatically the solution. Apply worthwhile non-blocking suggestions only when requested or clearly beneficial.

Keep detailed finding dispositions in the review artifact or coordinator handoff. Add only a compact changelog line to the design, for example:

- 2026-07-28: Addressed review pass 2; clarified token ownership and updated AUTH-002.

Leave the document Draft for independent follow-up review.

Report

Return the design path, a brief summary of the decisions and story order, the validation result, and any user decision still required. Do not start issue synchronization or implementation.