writing-plans

Generates bite-sized TDD implementation plans from specs with review gates.

Updated May 21, 2026
One-click install
npx skills add https://github.com/CagesThrottleUs/private-ai-harness --skill writing-plans-cagesthrottleus
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: writing-plans
Source: https://github.com/CagesThrottleUs/private-ai-harness/tree/main/skills/writing-plans
Command: npx skills add https://github.com/CagesThrottleUs/private-ai-harness --skill writing-plans-cagesthrottleus

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning a spec into code without a plan leads to scope creep, inconsistent types across tasks, and vague steps that stall implementers. This Skill converts a spec into a complete, self-contained implementation plan that an engineer with zero codebase context can execute task by task. ## Core Features & Use Cases - Bite-Sized TDD Tasks: Decomposes work into 2-5 minute steps (write failing test, verify failure, implement, verify pass, commit) with exact file paths, complete code, and expected command output. - Mandatory Quality Gates: Enforces design-principles and Karpathy anti-pattern checks, sequence diagram and API contract gates, a self-review checklist, and a plan-reviewer subagent pass before execution. - Oracle-Independence for Critical Tasks: Splits auth/payment/PII tasks into an independent test-writing dispatch and an implementer dispatch to prevent tests that merely confirm the planned implementation. - Use Case: After brainstorming produces a spec for a new feature, invoke this Skill to produce .ai/<date>-<feature>/plans/plans-<feature>.md, then hand off to subagent-driven or inline execution. ## Quick Start Ask the AI to use the writing-plans skill to create an implementation plan from the spec at .ai/2026-08-13-feature-design/specs/specs-feature-design.md.

Frequently Asked Questions about writing-plans

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

FAQPage Schema
How do I write an implementation plan from a spec before coding?

Invoke the writing-plans skill with a spec file path. It produces a plan document with a header linking the spec, global constraints, and bite-sized TDD tasks containing exact file paths, complete code, and commands with expected output.

How to break a large feature spec into implementation tasks?

The skill first checks scope: if the spec spans multiple independent subsystems, it recommends separate plans per subsystem. Within a plan, tasks are right-sized so each ends with an independently testable deliverable a reviewer could approve or reject on its own.

What makes a plan task fail review in this workflow?

Plan failures include placeholders like TBD or 'add appropriate error handling', steps without actual code, references to undefined types, tasks touching more than two files without justification, and handler tasks for HTTP endpoints lacking an approved OpenAPI or proto spec reference.

When does the oracle-independence gate apply to a task?

It applies only to Critical-tier tasks such as auth, payment, PII, or breaking public API changes. A separate subagent writes failing tests from the spec alone before any implementation exists, and the implementer must make those tests pass without editing the assertions.

Can I execute the plan inline instead of using subagents?

Yes. After the plan-reviewer passes, the skill offers two options: subagent-driven development with a fresh subagent per task, or inline execution using the executing-plans skill with batched checkpoints for review.