⏳ This skill is pending AI review.
Scores will appear once the review pipeline completes.
five-step-method
Universal decision framework. MUST run before taking action on any request. Applies to: add feature, fix bug, implement, build, create, refactor, optimize, improve, deploy, configure, integrate, make decision, evaluate proposal, review feedback, plan strategy, design solution, choose approach, set priority, allocate resources, launch project, hire role, define process, write proposal, draft plan, pick tool, select vendor, approve budget, schedule meeting, define scope, set deadline, write policy
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
// README
Five-Step Work Method
If this skill helps your AI agents make better decisions, give it a ⭐ on GitHub — every star helps another team avoid over-engineering and scope creep.
An engineering decision framework for AI coding agents, based on Elon Musk's five-step methodology. Battle-tested through a real production deployment cycle.
What It Does
Teaches AI agents to evaluate requirements before acting — whether writing code, making product decisions, planning strategy, or designing processes. Prevents the most common failure: solving problems that don't exist, or solving real problems with unnecessary complexity.
The Five Steps
- Question — Is this actually needed?
- Delete — Remove what shouldn't exist
- Simplify — Find the minimum reliable solution
- Accelerate — Speed up what remains
- Automate — Only automate validated processes
Most decisions end at Step 3.
Install
Quick install (recommended) — one command
Auto-detects which AI agents you have and installs the skill into the right place. No clone, no manual file copying.
npx five-step-method
Want to target a specific agent only?
npx five-step-method claude-code # → ~/.claude/skills/five-step-method
npx five-step-method cursor # → ./.cursor/rules/five-step-method.md
npx five-step-method windsurf # → ./.windsurfrules
npx five-step-method cline # → ./.clinerules
npx five-step-method codex # → ./AGENTS.md
npx five-step-method --list # show all supported targets
The installer is idempotent — running it twice won't duplicate content. For Windsurf/Cline, it wraps the skill in <!-- five-step-method:begin/end --> markers so it doesn't clobber your existing rules.
Published as
five-step-methodon npm.
Manual install
If you'd rather copy files yourself:
Claude Code
git clone https://github.com/agidesigner/five-step-method-skill.git
cp -r five-step-method-skill/.claude/skills/five-step-method ~/.claude/skills/
OpenClaw
git clone https://github.com/agidesigner/five-step-method-skill.git
cp -r five-step-method-skill/skills/five-step-method /path/to/openclaw/skills/
Codex (OpenAI)
cp five-step-method-skill/AGENTS.md /path/to/your/project/AGENTS.md
Cursor / Windsurf / Cline
Copy the content of SKILL.md into your project's custom instructions file (.cursorrules, .windsurfrules, or agent system prompt).
Structure
five-step-method/
├── .claude/skills/five-step-method/
│ └── SKILL.md # Claude Code
├── skills/five-step-method/
│ └── SKILL.md # OpenClaw
├── AGENTS.md # Codex (same content)
├── skill.json # Manifest
└── README.md
Origin
Extracted from a session where an AI agent:
- Built an auto-detection system → caused 3 production 500 errors
- Force-pushed tags 4x → blocked every deployment
- Stacked 5 untested changes → each fix introduced a new bug
- Eventually rolled back everything
After applying this method, the same problems were solved with single-line config changes.
Key Insight
The best code is code you didn't write. The best feature is the one you deleted. The best config is one line in
.env.
Author
Ajin — @ajinpro
Founder @ Jogg.ai | Building AI for Marketing
Show Your Support
If the Five-Step Method has saved you from over-engineering, scope creep, or shipping the wrong feature, here's how you can help others find it:
- ⭐ Star this repo — the single biggest signal that helps other developers discover it on GitHub
- Share it with teammates who use Claude Code, Cursor, Windsurf, Cline, or Codex
- Use it in your own projects —
npx five-step-methodand tell people where you got it - Open an issue with feedback, bug reports, or new agent integrations
- Tweet about it and tag @ajinpro
Every star pushes this skill into more developers' workflows — and every prevented over-engineered feature is a small win for the whole industry.
Star History
License
MIT
// HOW IT'S BUILT
// FILES
2 files shipped with this skill. Inline text is the catalog copy. The upstream link is the source repository.
skills/five-step-method/SKILL.md9.8 KBc8741f7113eaupstream ↗
--- name: five-step-method description: "Universal decision framework. MUST run before taking action on any request. Applies to: add feature, fix bug, implement, build, create, refactor, optimize, improve, deploy, configure, integrate, make decision, evaluate proposal, review feedback, plan strategy, design solution, choose approach, set priority, allocate resources, launch project, hire role, define process, write proposal, draft plan, pick tool, select vendor, approve budget, schedule meeting, define scope, set deadline, write policy, change workflow, assess risk, handle complaint, resolve conflict. ZH: 加功能, 修bug, 实现, 开发, 创建, 重构, 优化, 改进, 部署, 配置, 集成, 做决策, 评估方案, 处理反馈, 制定策略, 设计方案, 定优先级, 分配资源, 立项, 招人, 定流程, 写方案, 选工具, 排期, 定范围, 改流程, 评估风险, 处理投诉. Five steps: 1) Question need 2) Delete unnecessary 3) Simplify 4) Accelerate 5) Automate last. Prevents over-engineering, complexity, scope creep." --- # Five-Step Work Method A universal decision framework that teaches you to evaluate **whether** to act before deciding **how** to act. Apply this to any decision — code, product, strategy, process, hiring, meetings, or resource allocation. ## When to Apply | Trigger | Action | |---------|--------| | New feature / product request | **Must apply** — Question need before designing | | External feedback (multiple points) | **Must apply** — Evaluate each point independently | | Bug fix that requires > 10 lines | **Must apply** — Question if the fix adds complexity | | Process / workflow change proposal | **Must apply** — Can we delete the process instead? | | New tool / vendor / hire decision | **Must apply** — Is the role/tool actually needed? | | Meeting or review request | **Must apply** — Can this be an async message? | | Strategy / roadmap planning | **Must apply** — Delete low-impact items first | | Configuration / deployment issue | **Must apply** — Ask if code change is even needed | | Single-line typo or obvious fix | **Skip** — Just fix it | | User gives explicit instructions | **Skip** — Follow instructions, then review | ## The Five Steps ### Step 1: Question the Requirement Ask: **"Is this actually needed? What happens if we don't do it?"** - Challenge every request, including your own instincts - Check if the problem solves itself (e.g., agent self-corrects after retry) - Distinguish "nice to have" from "actually broken" - Ask who benefits and how often - If only one user hit it once, it's probably not a product problem Decision matrix: | Frequency | Impact | Action | |-----------|--------|--------| | Common + High | Do it | Proceed to Step 2 | | Common + Low | Maybe | Proceed to Step 2 | | Rare + High | Record | Log for later, don't code now | | Rare + Low | Skip | Don't do it | ### Step 2: Delete **Remove what shouldn't exist before adding anything.** - Delete the feature, code, or process that created the problem - Delete options when a sensible default works - Delete fallbacks when the primary path is reliable - Delete auto-detection when one config line suffices - Delete backwards-compatibility shims for things nobody uses Rule: **If you're not sure whether something is needed, delete it.** Adding it back is faster than maintaining unnecessary complexity. ### Step 3: Simplify **Find the minimum solution that works reliably.** | Complex (avoid) | Simple (prefer) | |-----------------|-----------------| | 30-line auto-detection from headers | One line in `.env` | | Smart fallback chain A → B → C → D | Single source of truth | | Middleware + dispatcher + config | One function | | "Handle both cases" | Pick one, document the other | | Build a UI for configuration | Tell users to edit a file | | Add a parameter for flexibility | Choose the right default | Rule: **One config line beats 30 lines of detection code.** A clear error message beats a graceful fallback. ### Step 4: Accelerate Only optimize what survived Steps 1-3. - Don't optimize what shouldn't exist - Don't speed up a complex solution when a simple one exists - Measure before optimizing - Optimize the critical path, not edge cases ### Step 5: Automate Only automate what has been manually validated. - Never automate a process you haven't done manually first - Never automate something that happens once - Never automate before simplifying - Automation is the LAST step, never the first ## Analysis Output Format When analyzing a requirement or feedback: ``` ### Step 1: Question [Is it needed? Who benefits? How often?] ### Step 2: Delete [What can we remove instead of adding?] ### Step 3: Simplify [What's the minimum reliable solution?] ### Conclusion Do / Don't do / Record for later — [one sentence reason] ``` Most analyses end at Step 3. Only proceed to Steps 4-5 for performance-critical or high-frequency operations. ## Feedback Batch Analysis When receiving multiple feedback points at once: 1. Number each point 2. Run Step 1 on each: Is it our problem? Is it actually broken? 3. Build a verdict table: ``` | # | Point | Verdict | Reason | |---|-------|---------|--------| | 1 | ... | Don't do | Agent self-corrects | | 2 | ... | Do | Confirmed bug, simple fix | | 3 | ... | Not ours | Client-side config issue | | 4 | ... | Record | Valid but low priority | ``` 4. Only items that survive Step 1 get Steps 2-3 5. Typical result: 5 feedback points → 1-2 actionable items ## Anti-Pattern Detection Catch yourself when you see these patterns: | What you're about to do | Ask yourself | |--------------------------|-------------| | Add auto-detection / smart defaults | Is one config line really too hard for the user? | | Add a fallback for edge cases | Does this edge case actually happen? | | Build it now for future flexibility | Has anyone asked for this? | | Add a parameter / option / flag | Can I just pick the right default? | | Fix a data issue with code logic | Should the user fix their data instead? | | Handle multiple cases in one function | Can I just handle the common case? | | Add error recovery / retry logic | Does the caller already retry? | | Write 5 lines "just in case" | Will these 5 lines need maintaining forever? | | Schedule a recurring meeting | Can this be async or a shared document? | | Add a review/approval step | Am I reacting to one incident with permanent process? | | Hire for a new role | Can I simplify the work so it doesn't need a new person? | | Adopt a new tool or platform | Can I fix how we use the current one? | | Create a new Slack channel / group | Does this reduce communication or add noise? | | Write a policy document | Can a one-line rule replace a 10-page policy? | ## Real-World Examples ### "Auto-detect public URL from request headers" - **Step 1:** Users deploy behind nginx, need correct URLs → Valid need - **Step 2:** Delete the detection — it has edge cases (MCP transport security, proxy headers) - **Step 3:** `APP_URL=https://domain` in `.env` → one line, zero edge cases - **Result:** Don't auto-detect. The "smart" solution caused 3 production incidents. ### "Agent passes merchant name instead of UUID" - **Step 1:** Agent self-corrects in 2 seconds by calling list_merchants → Works already - **Result:** Don't do. Record as low-priority ergonomic improvement. ### "Add example outputs to settings page" - **Step 1:** Settings page is for configuration, not product demos - **Step 2:** Delete the idea — wrong place for it - **Result:** Don't do. Settings page stays focused. ### "Script writer doesn't work for recruitment content" - **Step 1:** Product is designed for marketing scripts, not HR content - **Result:** Don't do. Not a bug — it's a scope boundary. ### "Force authentication token for public deployment" - **Step 1:** Self-hosted product, user decides their security posture - **Result:** Don't do. Design choice, not a deficiency. ### "We need a weekly status meeting for the new project" - **Step 1:** What decision requires this meeting? Can it be async? - **Step 2:** Delete the meeting. Replace with a shared doc updated async. - **Step 3:** If sync is truly needed, make it 15 min biweekly, not 60 min weekly. - **Result:** Delete or simplify. Most status meetings exist because no one questioned them. ### "Let's hire a dedicated DevOps engineer" - **Step 1:** What are they going to do daily? Is current CI/CD actually broken? - **Step 2:** Delete manual deployment steps first — automate with a shell script. - **Step 3:** If 2 hours/week of deployment work remains, it doesn't justify a full hire. - **Result:** Simplify the process first. Hire only if workload justifies after simplification. ### "We should switch from Tool A to Tool B" - **Step 1:** What specific problem does Tool A fail at? Is it the tool or how we use it? - **Step 2:** Can we fix Tool A's config instead of migrating? - **Step 3:** If migration is needed, what's the smallest scope that solves the actual problem? - **Result:** Fix before replacing. Migration cost is almost always underestimated. ### "Add a review/approval step to the publishing workflow" - **Step 1:** What went wrong that triggered this request? How often? - **Step 2:** If it was one incident, delete the proposal. Fix the root cause instead. - **Step 3:** If recurring, add a checklist instead of a human bottleneck. - **Result:** Don't add process to prevent rare events. Process has compounding cost. ## Pre-Action Checklist Before taking any action, verify: - [ ] I questioned whether this is needed (Step 1) - [ ] I considered what to remove instead of add (Step 2) - [ ] This is the simplest approach that works (Step 3) - [ ] I'm not adding process to fix a one-time incident - [ ] I'm not building for a hypothetical future - [ ] I'm not adding a tool/role/meeting when simplifying existing ones works - [ ] For code: I verified the change compiles/imports before shipping - [ ] For code: I'll ship → verify → then start the next changeREADME.mdupstream ↗
Contents are not inlined in the catalog. Open the upstream file.
Plain-text copy, no login: /skill-bundles/agidesigner-five-step-method-skill-skills-five-step-method.md