task

Decides whether a coding task needs a plan and routes it to planning or direct implementation.

Updated Nov 17, 2025
One-click install
npx skills add https://github.com/vzakharov/vovazakharov.com --skill task-vzakharov
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: task
Source: https://github.com/vzakharov/vovazakharov.com/tree/main/.claude/skills/task
Command: npx skills add https://github.com/vzakharov/vovazakharov.com --skill task-vzakharov

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When an AI coding agent receives a task, it must decide whether to plan first or implement directly, and whether the operator needs to approve the plan before work begins. This Skill makes that triage decision explicit and consistent, preventing both over-planning trivial changes and under-planning risky ones. ## Core Features & Use Cases - Two-gate triage: Step 1 decides if the task gets a written plan; Step 3 decides if that plan blocks on operator approval, judged against the written plan rather than the original request. - Three defined outcomes: plan and hand off, go directly (the diff is the plan), or plan then go with a conditional go-ahead recorded in a commit. - Draft-based handoff: every planning route produces a draft plan file published via PR, so a dead session can be resumed by flipping the draft. - Use Case: A developer asks the agent to refactor a shared module. The skill writes a draft plan, detects that a fork carries no recommendation, and stops for operator approval instead of guessing. ## Quick Start Invoke /task followed by a description of the change you want, and the agent will report whether it is planning, handing off, or implementing directly.

Frequently Asked Questions about task

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

FAQPage Schema
How do I decide whether a coding task needs a plan before implementing?

Write a plan when the change has ordered parts, a live reuse decision, or a shape you would otherwise discover mid-build. If you can hold the whole change in your head and name every file it touches, implement directly and let the diff be the plan.

When should an AI agent stop and wait for operator approval?

Stop when the work costs far more to produce than to describe, when a fork carries no recommendation, when a review round would come too late, or when the scope itself is the open question. Any one of these signs means the plan blocks on the operator.

What happens if a session dies after writing a plan but before approval?

Every planning route produces a draft plan file published through a PR, so a later session can read the draft and flip it to resume. The draft is the recovery surface, costing one round trip to re-give the go-ahead.

Can I override the agent's decision to skip planning?

Yes. The agent reports its call in the first sentence with an override option; replying with the word plan makes it stop, write a plan for the whole task, and name any commits already made without reverting them.

What are the limitations of conditional go-ahead authorization?

The go-ahead is scoped to the single task in the invoking message and does not extend to other work. Close calls break toward writing a plan and toward operator review, so trivial tasks may still produce a draft file that later gets swept.