| name | sequential-thinking-default |
| description | Use structured sequential thinking for multi-step debugging, architecture decisions, or ambiguous problem framing. |
Structured Sequential Thinking
When to use this skill
Use this skill when you need to:
- Debug failures where the root cause isn't obvious
- Compare multiple valid implementation approaches
- Plan multi-file changes without missing consumers
- Make architectural decisions with competing tradeoffs
- Explore solution spaces before committing to implementation
Core Protocol
See .kilocode/rules/general-workflow.md for the full protocol. Key rules:
- Granularity: One
process_thought call = one reasoning step, ≤4 sentences
- Stage purity:
stage field must match the thought's content
- Stage coverage: ≥3 distinct stages before entering Conclusion
- Session hygiene:
clear_history → think → generate_summary → export_session
- Metadata required: At least one of
tags, axioms_used, or assumptions_challenged per thought
Stage Progression
- Problem Definition (≥1 thought) — What are we solving?
- Research (≥1 thought) — What information do we need?
- Analysis (≥2 thoughts) — Compare approaches, evaluate tradeoffs
- Synthesis (≥1 thought) — How do pieces connect? Call
generate_summary here.
- Conclusion (≥1 thought) — Decision with rationale
Minimum: 6 thoughts across ≥3 stages for non-trivial tasks.
Thinking Styles (Composable Contracts)
For specialized reasoning, use a thinking-style contract:
Handoff format: thinking_handoff.md
Thinking Plans (Delegation Recipes)
For multi-style reasoning, orchestrators use thinking plan templates:
Tools
process_thought — Record stage-based thoughts with metadata
generate_summary — Retrieve session overview (mandatory before Conclusion)
export_session — Save session for future reference (mandatory at session end)
import_session — Load previous session to resume work
clear_history — Clear thought history (mandatory at new session start)
Integration with other tools
- Pair with codebase-retrieval: Use Research stage to plan information gathering, then execute retrieval
- Complement with Context7: When assumptions involve library APIs, verify with up-to-date docs
- Before major refactors: Use Analysis/Synthesis stages to evaluate approaches
Anti-Pattern: The Big Blob
Do NOT dump all reasoning into one massive thought. Each thought = one discrete step.
See general-workflow.md for correct vs incorrect examples.