spec

Converts vague feature intent into a precise, executable GitHub issue spec.

1|Updated Jul 23, 2026
One-click install
npx skills add https://github.com/sanjanb/my-agent-harness --skill spec-sanjanb
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: spec
Source: https://github.com/sanjanb/my-agent-harness/tree/main/skills/gstack-spec
Command: npx skills add https://github.com/sanjanb/my-agent-harness --skill spec-sanjanb

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning a rough idea or verbal request into a well-structured, backlog-ready specification is time-consuming and often produces ambiguous tickets. This Skill guides the agent through a five-phase process that files a GitHub issue, optionally spawns a Claude Code agent in a fresh git worktree to implement it, and lets the /ship workflow close the source issue on merge. ## Core Features & Use Cases - Five-phase spec generation: Transforms vague intent into a precise, executable specification with structured decision briefs at each step. - GitHub issue filing: Creates the issue directly so it lands in the project backlog, and integrates with /ship to auto-close it on merge. - Worktree agent spawning: Optionally launches a Claude Code agent in an isolated git worktree to execute the spec without disturbing the current branch. - Use Case: You say "spec out dark mode support for the settings page" — the Skill interviews you with structured decision questions, writes the spec, files it as a GitHub issue, and can hand it off to an implementation agent. ## Quick Start Ask the agent to spec out your feature idea and file it as a GitHub issue, for example by saying "turn the offline-sync idea into a backlog item".

Frequently Asked Questions about spec

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

FAQPage Schema
How do I turn a feature idea into a GitHub issue with an AI agent?

Invoke the spec skill with your idea in natural language, such as "spec this out" or "make this a GitHub issue". It runs a five-phase interview with structured decision questions, then files a backlog-ready issue on GitHub.

How to automatically implement a spec after writing it?

The spec skill can optionally spawn a Claude Code agent in a fresh git worktree to execute the spec. The implementation runs isolated from your current branch, and the /ship skill closes the source issue when the work merges.

Does the spec skill work in headless or CI environments?

Yes, it detects the session kind (interactive, spawned, or headless) via a preamble script. In headless sessions it blocks on decisions that need human input; in spawned sessions it auto-chooses recommended options and reports results as prose.

What happens if the interactive question tool is unavailable?

The skill falls back to rendering decision briefs as structured prose with an ELI10 explanation, per-option completeness scores, and a recommendation, then stops and waits for a typed reply. It never silently auto-decides outside spawned sessions.

When should I not use the spec skill?

Skip it for trivial changes that need no ticket, or when the work is already fully specified. It is designed for turning ambiguous intent into tracked backlog items, not for direct code edits or bug investigation, which other gstack skills handle.