# Super Plan Mode > This skill should be used only when the user explicitly mentions "super plan mode", "spm", "use super plan", "run super plan mode", "activate super plan", or directly invokes /super-plan-mode or /spm. Do NOT trigger for generic planning phrases like "plan this", "show me the plan", "think through the approach", or "plan before coding" — those should be handled by the agent's built-in plan mode. - Skill: `temikeezy/super-plan-mode` (Agent Skill, multi-file: 3 files) - Install (CLI): `npx skillmds@latest add temikeezy/super-plan-mode` - Raw SKILL.md: https://api.skillmd.com/api/skills/temikeezy/super-plan-mode/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - License: MIT - Author: temikeezy (https://skillmd.com/u/temikeezy) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/temikeezy/super-plan-mode --- # Super Plan Mode Skill This skill activates Super Plan Mode: a structured planning workflow that generates a rich implementation plan and halts for explicit user approval before any code changes are made. ## Behavior When Active When this skill is active: 1. DO NOT create, edit, or delete any files until the user has explicitly approved the plan. 2. Announce plan mode immediately: "Entering super plan mode. I will generate a plan and wait for your approval before making any changes." 3. Run the full planning workflow described below. 4. Present the numbered acceptance gate and wait. ## Compatibility This skill works in any agent environment — Claude Code, Codex CLI, Cursor Agent, Windsurf, Aider, or any text-based AI coding assistant. See `references/agent-compat.md` for environment-specific notes. ## Planning Workflow ### Step 1: Clarify (if needed) If the request is ambiguous, ask 1–3 focused questions before exploring. If the request is clear, proceed directly to exploration. Do not ask more than 3 questions in a single message. Start with the most important. ### Step 2: Explore the Codebase (Read-Only) **If subagents are supported (OpenClaw, Claude Code, Cursor Agent, Windsurf Cascade, Codex CLI):** Launch 2–3 `plan-researcher` agents in parallel, each with a distinct focus: - Agent A: architecture mapping and existing patterns - Agent B: affected files and dependency tracing - Agent C: risks, uncommitted changes, and fragile areas After agents return, read all files they identify as critical. **If no subagent support (Aider or unknown environments):** Perform exploration inline sequentially following the same methodology: 1. Map architecture and patterns (Glob, Grep, Read) 2. Trace affected files and dependencies 3. Assess risks — run `git status` and `git log --oneline -10` In both cases: read critical files, understand existing patterns, and identify risks before generating the plan. ### Step 3: Generate the Plan Produce the plan document following the format in `references/plan-template.md` exactly. **Before generating, check for `.super-plan-mode.json`** in the project root. If present, read it and apply its defaults (planSaveDir, autoPhaseCheckpoints, etc.). Save the plan to: `[planSaveDir]/super-plan-mode-[unix-timestamp].md` Default save location: `.claude/plans/super-plan-mode-[timestamp].md` **MANDATORY: After saving, output this line verbatim (with the actual path filled in):** ``` 📁 Plan saved to: `.claude/plans/super-plan-mode-[timestamp].md` ``` This must appear before the acceptance gate. Never skip it. ### Step 4: Present the Acceptance Gate **STOP. After presenting the plan and the save location, you MUST output the gate below. Do NOT proceed to implementation, do NOT write any files, do NOT take any action until the user selects an option.** Output the numbered menu exactly as follows: ``` ──────────────────────────────────────────── Plan ready. What would you like to do? 1 Accept & Build 2 Reject / Cancel 3 Modify plan 4 Accept Phase 1 only ← (shown only for L/XL estimates) ──────────────────────────────────────────── Enter number (or type the action): ``` **HALT HERE. Do not write a single file or run any command until the user replies.** Response handling — execute immediately upon receipt: - **1 / "accept" / "build" / "yes" / "y"** → proceed to implementation (Step 5) - **2 / "reject" / "cancel" / "no" / "n"** → acknowledge, summarize useful research findings, stop; do not modify any files - **3 / "modify" / "m"** → ask what to change, revise the plan, show a diff of changed sections, re-present gate - **4 / "phase 1" / "p1"** → implement only Phase 1 steps, then pause and re-present gate for remaining phases ### Step 5: Implement (ONLY After User Selects Option 1 or Option 4) **Do not begin this step unless the user replied "1", "accept", "build", "yes", "y", "4", "phase 1", or "p1" to the gate.** Execute plan steps in order. For L/XL estimates with phases: after each phase completes, pause and report: - What was completed - What comes next - Any unexpected findings that may affect the remaining plan Ask the user to confirm before starting the next phase. On full completion, summarize: files created/modified, steps completed, deviations from plan (if any), suggested next steps. ## Additional Resources - **`references/plan-template.md`** — canonical plan output format with all required sections - **`references/agent-compat.md`** — per-agent environment notes for exploration, gate behavior, model flags, and plan file handling