# Chain Of Thought > Explicit step-by-step reasoning for complex problems including multi-step analysis, architecture decisions, and debugging. Use when showing the reasoning process is as valuable as the answer. - Skill: `marktantongco/chain-of-thought` (Agent Skill) - Install (CLI): `npx skillmds@latest add marktantongco/chain-of-thought` - Raw SKILL.md: https://api.skillmd.com/api/skills/marktantongco/chain-of-thought/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: marktantongco (https://skillmd.com/u/marktantongco) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/marktantongco/chain-of-thought --- # Chain-of-Thought Reasoning ## Context Activate this skill when the user presents a **complex problem** that benefits from explicit, step-by-step reasoning before arriving at a conclusion. This includes: - Multi-step mathematical, logical, or analytical problems - Architecture or design decisions with many moving parts - Debugging scenarios requiring causal reasoning - Any task where showing the reasoning process is as valuable as the answer - Situations where the agent might otherwise jump to a wrong conclusion by skipping steps **Do not use** for simple factual lookups, greetings, or tasks where the user only wants a direct answer. ## Instructions ### Step 1: Restate the Problem Paraphrase the problem in your own words to confirm understanding. Identify what is being asked, what is given, and what the expected output looks like. ### Step 2: Identify and State Assumptions List every assumption you are making before reasoning begins. Number them explicitly. If any assumption is uncertain, flag it with `[ASSUMPTION — UNVERIFIED]`. ### Step 3: Decompose into Sub-Problems Break the problem into the smallest discrete steps possible. Each step should be independently verifiable. Assign a number to each step. ### Step 4: Solve Each Step Explicitly For every sub-problem: - State what you are calculating or reasoning about - Show the work (equations, logic, references) - State the intermediate result - Verify the result makes sense before moving on ### Step 5: Cross-Check and Validate After completing all steps, revisit the chain: - Do the intermediate results chain together correctly? - Does the final answer satisfy the original problem statement? - Run a quick sanity check (e.g., back-of-the-envelope, boundary test) ### Step 6: Flag Uncertainties After presenting the final answer, list any remaining uncertainties, caveats, or areas where assumptions could change the result. Rate confidence as High / Medium / Low for each. ## Constraints 1. **Never skip steps.** Every logical jump must be explicit. If you find yourself wanting to skip, add a step explaining why. 2. **Number every step.** Use `Step N:` format consistently. 3. **Assumptions come first.** Never reason from hidden assumptions — state them upfront. 4. **Show your work.** Intermediate calculations, logical derivations, and references must be visible. 5. **One conclusion per chain.** If the problem branches, start a new numbered chain for each branch. 6. **Flag, don't hide.** If you are unsure about a step, mark it clearly rather than presenting it as confident. 7. **No premature answers.** The final answer appears only after Step 5 is complete. ## Examples ### Example 1: Technical Architecture Decision **User:** "We need to choose between a monolith and microservices for our e-commerce platform with 50k daily users." **Output:** ``` Chain-of-Thought Reasoning Step 1: Restate the Problem We must choose between a monolithic architecture and microservices for an e-commerce platform serving approximately 50,000 daily active users. The decision should optimize for reliability, developer velocity, and future scalability. Step 2: Assumptions 1. [ASSUMPTION] The team size is small-to-medium (3–10 developers). 2. [ASSUMPTION] Traffic growth is expected to be steady, not exponential. 3. [ASSUMPTION — UNVERIFIED] The platform includes checkout, inventory, user accounts, and a recommendation engine. Step 3: Decompose 3a. Evaluate monolith against our criteria 3b. Evaluate microservices against our criteria 3c. Compare at current scale (50k DAU) 3d. Compare at projected 2x scale 3e. Consider team-structure alignment Step 4: Solve 4a. Monolith: simpler deployment, single database, faster initial dev... 4b. Microservices: independent scaling per service, fault isolation... 4c. At 50k DAU, a well-built monolith handles this comfortably... 4d. At 100k DAU, checkout may need independent scaling... Step 5: Cross-Check The monolith is the stronger fit at current scale. Microservices become justified when a specific service hits a scaling ceiling. Step 6: Uncertainties - [Medium confidence] Assumption #1 about team size — if the team is larger than 15, microservices gain advantage. - [Low confidence] Future traffic projections are not confirmed. Recommendation: Start with a modular monolith... ``` ### Example 2: Mathematical Problem **User:** "A factory produces 400 units/day. Defect rate is 3%. Each defective unit costs $15 to rework. What is the monthly rework cost?" **Output:** ``` Chain-of-Thought Reasoning Step 1: Restate the Problem Calculate total monthly cost of reworking defective units given a daily production of 400 units with a 3% defect rate and a $15 rework cost per defective unit. Step 2: Assumptions 1. The factory operates 7 days a week (30-day month). 2. Defect rate is consistent and independent of production volume. Step 3: Decompose 3a. Calculate daily defective units 3b. Calculate daily rework cost 3c. Calculate monthly rework cost Step 4: Solve 4a. Daily defective units = 400 × 0.03 = 12 units 4b. Daily rework cost = 12 × $15 = $180 4c. Monthly rework cost = $180 × 30 = $5,400 Step 5: Cross-Check Sanity: 3% of 400 is ~12. 12 × 30 = 360 defective/month. 360 × $15 = $5,400. ✓ Matches. Step 6: Uncertainties - [Low confidence] Operating days assumption — if 5-day work week, result drops to ~$3,900/month. Final Answer: $5,400/month under 7-day operations. ```