stark-author

Authors human-gated specs and task DAGs through recon, interview, and sign-off.

Updated Mar 16, 2026
One-click install
npx skills add https://github.com/21StarkCom/stark-skills --skill stark-author-21starkcom
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: stark-author
Source: https://github.com/21StarkCom/stark-skills/tree/main/runtime-overrides/codex/skill/stark-author
Command: npx skills add https://github.com/21StarkCom/stark-skills --skill stark-author-21starkcom

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Writing a spec that an autonomous implementer can execute without re-planning is hard: vague done-when criteria, missing wiring seams, and unverified file references cause implementation failures. This Skill runs a single structured session that produces a self-contained, machine-checkable spec+plan the operator signs off on before any code is written. ## Core Features & Use Cases - Tiered authoring workflow: Classifies the change as Skip, Short, or Full, then runs time-boxed repo recon, a plain-language operator interview, and spec authoring with EARS-patterned behavior criteria and a task DAG. - Built-in QA gates: Enforces wiring-seam checks, rejects vacuous done-when commands, runs one zero-context advisory pass, and produces a plain-English .human.md sidecar for non-technical sign-off. - Commit-pinned handoff: Stamps the accepted base commit, opens a draft PR on a spec/<slug> branch, and prints the implementer's first-move contract. - Use Case: Before starting a multi-file feature, invoke the skill with your intent; it reads the code itself, asks you only product questions, and delivers a signed-off spec ready for autonomous implementation. ## Quick Start Invoke the skill with a short description of the change you want, for example: author a spec for adding rate limiting to the public API endpoints.

Frequently Asked Questions about stark-author

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

FAQPage Schema
How do I write a spec for autonomous AI code implementation?▼

Invoke the skill with your intent or a notes path. It performs time-boxed repo recon, asks only product questions you alone can answer, then writes a self-contained spec with EARS behavior criteria, a task DAG, and machine-checkable done-when commands for each task.

What makes a done-when criterion machine-checkable in a spec?▼

A valid done-when is a command that fails if the work was never done, such as a test naming the exact artifact. The skill rejects vacuous checks like test filters matching zero tests, multi-file greps that pass on partial matches, or build commands standing in for behavior checks.

When should I use the Skip, Short, or Full tier?▼

Skip applies when the diff is describable in one sentence. Short fits a single-area, low-uncertainty change needing a 10-30 line doc. Full applies when the approach is uncertain, multiple files are involved, or the code is unfamiliar, requiring the complete template and interview.

Does the spec review use LLM review loops?▼

No. The skill runs exactly one zero-context advisory pass per body of text, triages findings once, and never re-reviews text it already saw. Only genuinely new sections added after sign-off feedback get a single scoped first-pass review.

What happens after the operator approves the spec?▼

The skill stamps the accepted base commit hash, creates a spec/<slug> branch, commits the spec and its plain-English sidecar, opens a draft PR, and prints the implementer's first-move contract for the handoff to the build stage.