sequential-thinking

# Sequential Thinking A reasoning harness around the `mcp__sequential-thinking__sequentialthinking` tool. The MCP tool provides persistence, revision, and branching for thoughts; this skill provides the *discipline* for using it well — when to invoke it, how to structure each thought, which reasoning mode fits which problem, how to ground claims against hallucination, when to challenge yourself adversarially, and when to stop. The core insight: smart reasoning isn't about thinking harder. It's about thinking with the right mode for the situation, anchoring claims to evidence, and knowing when one approach isn't enough. --- ## Decide whether to use this skill at all Not every question needs structured reasoning. Activating this skill on a trivial question wastes tokens, slows the response, and trains a reflex of overthinking. Skipping it on a hard question produces shallow answers and hallucinations. The cost of misuse cuts both ways. Two tests decide. Use the skill if either fires: **Reversibility test.** Is the consequence of being wrong easy to undo, or hard? - *Easy to undo* (Type 2 / two-way door): pick a variable name, suggest a library, draft prose, fix an obvious typo. Skip the skill. A mistake costs seconds. - *Hard to undo* (Type 1 / one-way door): choose an architecture, ship code to production, commit to a design that other code will depend on, advise on an irreversible decision. Activate the skill. A mistake costs hours, days, or trust. **System-2 signal t...