Sequential Thinking

---
name: mk:sequential-thinking
version: 1.0.0
description: |
  Use when a problem requires structured step-by-step reasoning before acting.
  Prevents jumping to conclusions or guessing root causes without evidence.
  Triggers on complex debugging, multi-step analysis, hypothesis verification,
  or when mk:fix invokes diagnosis phase. NOT for being stuck on approach
  (too many implementations, forced assumptions, innovation block) — use
  mk:problem-solving instead.
allowed-tools:
  - Read
  - Grep
  - Glob
  - Bash
source: local
keywords: [sequential-thinking, step-by-step-reasoning, mcp-sequentialthinking, evidence-based, incremental-analysis, chain-of-thought]
when_to_use: "Use when applying step-by-step evidence-based reasoning. NOT for systematic debugging (see mk:investigate)."
user-invocable: true
---

# Sequential Thinking

Structured step-by-step reasoning with hypothesis generation, elimination, and revision capability.

## When to Use

- Complex problem requiring multi-step reasoning
- Debugging where root cause isn't obvious
- When mk:fix invokes diagnosis (Phase 3)
- Architecture decisions with multiple trade-offs
- Any situation where "I think it's X" needs evidence before acting

**Do NOT use for:** trivial fixes (typo, lint), single-file changes with obvious cause.

## Process

1. **State the problem** — what is observed vs expected
2. **Generate hypotheses** — from evidence, NOT guessing. Each hypothesis needs: what evidence would confirm it, what would refute it
3. **Test hypotheses** — use Grep/Read/Bash to find evidence for/against each
4. **Eliminate** — mark each CONFIRMED, REFUTED, or INCONCLUSIVE with evidence
5. **Conclude** — state root cause with confidence level + evidence chain
6. **Scope the fix** — what must change to address root cause (not symptoms)

Adjust dynamically: expand if more complex than expected, contract if simpler.

## Output Format

Use `references/hypothesis-testing.md` template. Key sections:

```markdown
## Sequential Analysis: [problem]

### Hypotheses

| #   | Hypothesis | Evidence for | Evidence against | Status |
| --- | ---------- | ------------ | ---------------- | ------ |

### Root Cause

Confirmed: [root cause]
Confidence: [high/medium/low]
Evidence: [what confirms this]

### Fix Scope

[what must change — files, functions, logic]
```

## Revision & Branching

When new evidence invalidates a previous conclusion:

- Mark revision explicitly: `[REVISION of Hypothesis N]: [correction + why]`
- When 2+ approaches viable: branch, evaluate each, converge with decision

Load `references/core-patterns.md` for revision/branching patterns.
For advanced techniques (spiral refinement, multi-branch convergence): `references/advanced-techniques.md`
For uncertainty + revision cascades: `references/advanced-strategies.md`
For concrete examples: `references/examples-debug.md`, `examples-api.md`, `examples-architecture.md`

## Diagnostic Frameworks

Load when a specific methodology fits the investigation:

- `references/five-whys-plus.md` — enhanced 5-Whys with bias guards + stopping criteria. Use for post-mortems, recurring problems, human-error investigations.
- `references/scientific-method.md` — hypothesis → falsifiable prediction → experiment → revise. Use for A/B tests, performance investigations, production incidents.
- `references/kepner-tregoe.md` — SA/PA/DA/PPA + IS/IS-NOT matrix. Use for multi-system bugs, contested root causes, high-stakes decisions.

## Scripts

Track and format thoughts deterministically:

- `scripts/process-thought.js` — validate, track history, branches, revisions. Max 20 thoughts.
  `node scripts/process-thought.js --thought "..." --number N --total M --next true`
  `node scripts/process-thought.js --summary` (context-efficient handoff to mk:fix)
- `scripts/format-thought.js` — format as box/simple/markdown/json
  `node scripts/format-thought.js --thought "..." --number N --total M --format json`

## Gotchas

- **Premature conclusion**: confirming first hypothesis without testing alternatives → test at least 2 hypotheses before concluding
- **Evidence-free hypotheses**: "I think it's X" without evidence → every hypothesis needs at least one piece of evidence
- **Revision avoidance**: refusing to revise when new evidence contradicts → revise explicitly, don't force-fit

Full list: `references/gotchas.md`

## Workflow Integration

Called by `mk:fix` during diagnosis (Step 2). Receives input from `mk:investigate` or direct observation. Output feeds into fix implementation.

```
mk:investigate → symptoms + traces → mk:sequential-thinking → root cause → mk:fix
```