kiro-discovery

Determine optimal action paths for work items by scanning repository metadata and steering artifacts.

Updated Apr 16, 2026
One-click install
npx skills add https://github.com/manatago/gion --skill kiro-discovery-manatago
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: kiro-discovery
Source: https://github.com/manatago/gion/tree/main/.claude/skills/kiro-discovery
Command: npx skills add https://github.com/manatago/gion --skill kiro-discovery-manatago

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Kiro-discovery helps teams decide the correct next action for a proposed piece of work by quickly scanning repository metadata and steering artifacts to determine whether to update an existing spec, create a new spec, decompose into multiple specs, or proceed with direct implementation. It prevents wasted effort on unnecessary specification, keeps discovery conversations focused, and ensures the next step is actionable.

Core Features & Use Cases

  • Lightweight scan: Collects only metadata (file existence and spec.json name/phase/approvals) and a top-level directory listing to classify the project's state without loading full file contents.
  • Action path decisioning: Applies clear rules for Path A (existing spec), Path B (no spec), Path C (single new spec), Path D (multi-scope decomposition), and Path E (mixed decomposition) and prompts for confirmation when needed.
  • Controlled deep-loading & delegation: For non-trivial cases, it reads steering docs and relevant specs and delegates heavy codebase exploration to subagents to keep main context under budget.
  • Authoritative disk writes: For spec creation paths, it writes brief.md and roadmap.md files in the prescribed formats and verifies they exist so discovery results persist across sessions.
  • Use cases: onboarding a new feature request, triaging cross-cutting change proposals, and converting stakeholder ideas into scoped briefs and roadmaps ready for implementation.

Quick Start

Use kiro-discovery to evaluate a new feature request and recommend whether to update an existing spec, create one, decompose into multiple specs, or implement directly.

Frequently Asked Questions about kiro-discovery

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

FAQPage Schema
How do I decide whether to update an existing spec or create a new one for a feature request?

Spec discovery works by scanning repository metadata and steering artifacts to classify project state, then applying rules to choose between updating an existing spec, creating a new one, decomposing into multiple specs, or implementing directly.

What's the best way to decompose a cross-cutting change proposal into actionable specs?

Roadmap decomposition is handled by evaluating the proposed work against existing spec.json files and steering docs, then routing multi-scope changes through targeted decomposition paths that write brief.md and roadmap.md files for each scope.

How does spec-writing discovery keep the main context under a size budget?

Context size budgeting is maintained by performing lightweight metadata scans initially, then delegating heavy codebase exploration to subagents and selectively reading only relevant spec.json and roadmap.md files when deeper analysis is required.

When do I need to write brief.md and roadmap.md files during product discovery?

Writing brief.md and roadmap.md files is required when discovery routes to spec creation paths, ensuring that scoping decisions and action paths persist on disk across sessions for subsequent implementation.

Can I use this discovery workflow to triage stakeholder ideas without loading full repository contents?

Triage without loading full contents is supported through lightweight scans that collect only file existence and spec.json metadata, plus a top-level directory listing to classify the project state before any deep-loading occurs.

What limitations exist when dispatching subagents for codebase research during discovery?

Subagent dispatching is limited to non-trivial cases where steering docs and relevant specs require deep-loading, keeping the main context under budget by restricting subagent activity to targeted codebase exploration only.