director

Classifies human messages into seven acts and routes accepted work to the build pipeline.

1|Updated Jul 4, 2026
One-click install
npx skills add https://github.com/happysasha18/live-spec --skill director-happysasha18
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: director
Source: https://github.com/happysasha18/live-spec/tree/main/skills/director
Command: npx skills add https://github.com/happysasha18/live-spec --skill director-happysasha18

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? When working with an AI coding agent, casual messages often get misread as work requests, turning questions into tickets and ideas into roadmap rows. This Skill reads every human turn first, classifies what the person actually did, and routes only genuine work forward. ## Core Features & Use Cases - Act classification: Distinguishes seven acts (question, idea, observation, decision, correction, instruction, halt) by meaning and context rather than trigger words. - Route contract: Returns a structured verdict with acts, action, work item counts, dimensions, and specialists so downstream pipelines know exactly what to do. - Multi-act turns: Reads a whole message to the end and lists every act before deciding, preventing silently dropped requests. - Use Case: A user says "the onboarding copy feels stiff, and by the way deploy this one." The Director records the observation as evidence, accepts the deploy as one instruction, and creates exactly one new work item instead of guessing. ## Quick Start Ask the director skill to read and classify this message before any work begins on it.

Frequently Asked Questions about director

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

FAQPage Schema
How do I stop an AI agent from turning every message into a task?▼

Use a classification step that reads each message before acting on it. The Director classifies messages into seven acts and only creates work for genuine instructions, answering questions and recording observations without opening tasks.

How to route user requests to different AI specialist agents?▼

Classify the request first, then name the dimensions the accepted work touches and map them to specialists. The Director outputs a route contract listing the smallest set of specialists, such as architect, spec author, or test author, for the pipeline to call.

What is the difference between a correction and a new instruction?▼

A correction changes the goal or constraints of work already in flight and creates zero new work items, while an instruction asks for something new. The Director attaches corrections to the existing task instead of opening a duplicate row.

Can the Director handle messages with multiple requests at once?▼

Yes. It reads the entire turn to the end and lists every act before deciding, so a message containing a musing, a correction, and an instruction produces exactly one new work item without losing any act.

When should the Director ask a clarifying question instead of acting?▼

It asks one short question only when guessing wrong would change the result, phrased in the person's own terms. It never creates work to be safe, since creating unrequested work is the expensive failure direction.