plan

Writes implementation plans to git-tracked files published as draft PRs for review.

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

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Plan-mode UIs and AskUserQuestion dialogs lose answers in remote or web sessions, and plans presented in approval dialogs cannot be reviewed later as diffs. This Skill moves planning onto disk: the plan lives in a git-tracked file under docs/plans/ whose filename encodes its approval state, so it can be reviewed from any machine, commented on as a draft PR, and resumed in a later session. ## Core Features & Use Cases - On-disk plan files with an approval gate: Plans are written to docs/plans/<slug>.draft.do-not-implement.md and only renamed to .in-progress.md after an explicit operator go-ahead, making approval a recorded git event. - Draft PR publication and handoff: Every plan is published as a draft PR for inline review, and each planning turn ends with a copyable /go <branch> block so implementation starts in a fresh session from a recorded decision. - Numbered-prose questions and task carving: Clarifying questions are asked as numbered, lettered prose with recommended defaults, and oversized tasks can be carved into parent/child GitHub issues filed only on approval. - Use Case: In a remote Claude Code session, instead of entering plan mode, the agent researches the task, writes docs/plans/add-usage-charts.draft.do-not-implement.md with a DRY notes section, opens a draft PR, and ends the turn with a /go command the operator pastes into a new session after approving. ## Quick Start Ask the agent to plan a task using the plan skill so it writes the plan to docs/plans/ and publishes it as a draft PR instead of entering plan mode.

Frequently Asked Questions about plan

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

FAQPage Schema
How do I make an AI coding agent write plans to a file instead of plan mode?

Invoke the plan skill, which writes the plan to docs/plans/<branch-slug>.draft.do-not-implement.md, commits it, and publishes it as a draft PR. The filename's do-not-implement suffix blocks source edits until an explicit go-ahead renames it.

How do I review an agent's plan as a pull request diff?

The plan skill commits the plan file and invokes the pr skill to open a draft PR, so feedback arrives as inline comments and threads on the same surface used for code review. The PR URL is printed alongside the handoff block.

Why do Claude Code web sessions lose plan mode answers?

Web and remote sessions silently lose answers given to plan-mode and AskUserQuestion prompts that are re-emitted after the session idles. This skill avoids those UIs entirely by keeping plans and questions as durable git-tracked files and transcript prose.

When is it safe to start implementing after a plan is written?

Only when the operator's message literally contains a go-ahead token such as "go ahead", "implement", or "ship it". The first post-approval action is renaming the plan from .draft.do-not-implement.md to .in-progress.md, which is the single act that unlocks source edits.

When should a task be carved into multiple GitHub issues?

Carve only when work is genuinely too large for one PR, such as multiple unrelated subsystems or independently shippable deliverables. Each child issue should be substantial enough to justify its own PR, roughly five or more files of real change.