How to use the Strategic Planning Skill (omc-plan) for agents

What does the omc-plan skill make the agent do?

When the skill is installed, the agent plans before doing work. The skill picks a mode based on how specific your request is. A broad or vague request triggers an interview, where the agent asks you one question at a time about scope, constraints and preferences. A detailed request skips the interview and writes a plan directly.

The skill also supports two review styles. Consensus mode runs an iterative Planner, Architect and Critic loop. Review mode has the Critic evaluate a plan that already exists.

• Interview mode: for broad or vague requests, one question at a time
• Direct mode: for detailed requests, a plan is written immediately
• Consensus mode: Planner, Architect and Critic loop until they agree
• Review mode: the Critic evaluates a plan you already have

Install the omc-plan skill

Download the ZIP from this page and unzip it. The skill's only file is SKILL.md. Put the unzipped folder in one of two places: ~/.claude/skills/ to make it available everywhere, or a project's .claude/skills/ folder to limit it to that project.

You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/
• Claude apps: Settings → Capabilities → Skills, then upload

Check what the skill requires

The skill needs Claude Code or a compatible SKILL.md agent runtime. It relies on companion skills and an explore agent from the oh-my-claudecode framework, so those need to be available for the full workflow. Codex CLI is optional and only needed if you use the --architect codex or --critic codex overrides.

The skill runs no scripts and uses no network access, according to its listing.

• Required: Claude Code or a compatible SKILL.md runtime
• Required: companion skills and the explore agent from oh-my-claudecode
• Optional: Codex CLI, for --architect codex or --critic codex

How do you trigger the skill with natural requests?

You do not need a special command. Phrases such as "plan this", "let's plan" or "make a plan for" trigger planning. To ask for a review of an existing plan, say "review this plan" or use --review. For a multi-perspective pass, use --consensus or say "ralplan". Add --direct to skip the interview, --interactive to get prompts during consensus review, and --deliberate for a stricter consensus pass on high-risk work.

The argument hint for the skill is: [--direct|--consensus|--review] [--interactive] [--deliberate] <task description>

• "Plan this" or "let's plan": starts planning in the default mode
• "Review this plan" or --review: critiques an existing plan
• --consensus or "ralplan": runs the Planner, Architect and Critic loop
• --direct: skips the interview and writes a plan
• --interactive: adds user prompts at draft review and final approval in consensus mode
• --deliberate: uses deliberate consensus mode

Examples of planning requests

A vague idea works well with the interview. Ask something like "plan this: we need better search for our docs" and the agent will ask focused questions, looking at the codebase first where it can, before asking about patterns it could find itself.

A detailed task can go straight to a plan. Ask "plan the migration of the login form to the new validation module, with the existing tests kept passing" and the agent writes the plan directly.

For a second opinion, paste or point to an existing plan and say "review this plan". For a higher-stakes change, ask for consensus with --deliberate.

Understand approval and what is never run automatically

Plans are saved as files marked pending approval. Before you explicitly approve execution, planning modes do not run mutation-oriented shell commands, edit source files, commit, push, open PRs, invoke execution skills or delegate implementation. Plans are written to .omc/plans/ files matching ralplan-*.md when consensus mode is used.

Consensus mode is fully automated by default. Add --interactive if you want prompts at the draft and approval steps.

• Plans are marked pending approval until you approve
• No source edits, commits or pushes happen before explicit approval
• Consensus plans are saved as .omc/plans/ralplan-*.md

Tips for getting good results

Answer the agent's questions one at a time; the skill is designed to ask one question per turn. Let it look at the codebase before you answer questions about your own code. If a task is a single obvious fix, skip planning and use an execution skill directly.

Do not expect planning to write code. Use the plan as input for the next step, and when you are ready to execute, use the tool that fits the task, such as ralph or team, which the skill itself names as handoffs.

• Give a clear scope sentence to shorten the interview
• Use --review on plans written by someone else
• Use --direct when you already know what you want