ptp-writing-plans

Writes ordered, verifiable, agent-executable task checkboxes for a decided change.

Updated Jun 8, 2026
One-click install
npx skills add https://github.com/AlmogMaayan/ptp --skill ptp-writing-plans-almogmaayan
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ptp-writing-plans
Source: https://github.com/AlmogMaayan/ptp/tree/main/skills/ptp-writing-plans
Command: npx skills add https://github.com/AlmogMaayan/ptp --skill ptp-writing-plans-almogmaayan

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning a decided change proposal into an implementation plan often produces vague, unverifiable, or misordered tasks that an executing agent cannot complete unaided. This Skill converts a change's proposal, design, and spec deltas into a short, dependency-ordered tasks.md where every checkbox names exact files, an observable outcome, and an automated verification. ## Core Features & Use Cases - Verifiable checkbox authoring: Every checkbox names the exact files it creates or modifies, an observable outcome, and a verify: clause pointing to a test, command assertion, file-content check, automated browser test, or openspec validate. - Testability discipline: Behavior-changing checkboxes must name the specific test file and test case; prose-only edits carry a [prose-exempt: <reader>] marker; render-affecting work must name a browser-engine test. - Dependency ordering and terminal state: Checkboxes are ordered so each relies only on earlier ones, and changes too large for the budget return NEEDS SPLIT with a proposed decomposition into child changes. - Use Case: After deciding a change's proposal and design in an OpenSpec-based planning flow, run this Skill to produce openspec/changes/<change-id>/tasks.md that an implementing agent can execute and self-verify without human intervention. ## Quick Start Ask the agent to run the ptp-writing-plans skill on the current change to write its tasks.md implementation plan.

Frequently Asked Questions about ptp-writing-plans

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

FAQPage Schema
How do I write an implementation plan an AI agent can execute unaided?▼

Write one checkbox per deliverable that names the exact files it creates or modifies, the observable outcome, and a verify clause pointing to an automated check the agent can run itself. Order checkboxes so each depends only on earlier ones, and never include manual confirmation steps.

What makes a task checkbox verifiable in an automated plan?▼

A verifiable checkbox ends with a verify clause naming a test file and test case, a command plus an output assertion, a file-content assertion, an automated browser check, or an openspec validate call. Bare phrases like run the tests or manually confirm are not accepted.

How are prose-only documentation edits handled in a task plan?▼

A checkbox that changes no executable behavior, such as editing a SKILL.md or command file, carries the literal marker [prose-exempt: <reader it binds>] naming who consumes that prose, instead of naming a test file and case.

What happens when a change is too large for one plan?▼

The skill stops and returns NEEDS SPLIT with a proposed division into two or more child changes in dependency order, each with a one-line scope. It writes no further artifact text, and the caller treats the result as a decomposition instruction.

Can a plan include git commit or manual verification steps?▼

No. Git add, commit, push, and branch tasks are banned because committing is a separate explicitly requested step. Manual verification steps are also banned; unverifiable tasks must be reshaped, substituted, or relocated per the tasks-authoring capability.