Sequential Think Skill | Agent Skills

Sequential Think

Structured iterative thinking for complex problem-solving. Standalone CLI only (no MCP dependency).

Execution Methods

Run scripts/sequential_think_cli.py via Bash:

# Process a thought
python scripts/sequential_think_cli.py think \
  --thought "First, let me analyze the problem structure..." \
  --thought-number 1 \
  --total-thoughts 5

# Continue thinking chain
python scripts/sequential_think_cli.py think \
  --thought "Based on step 1, I hypothesize that..." \
  --thought-number 2 \
  --total-thoughts 5

# Revise a previous thought
python scripts/sequential_think_cli.py think \
  --thought "Reconsidering step 1, I realize..." \
  --thought-number 3 \
  --total-thoughts 5 \
  --is-revision \
  --revises-thought 1

# Branch into alternative path
python scripts/sequential_think_cli.py think \
  --thought "Alternative approach: what if we..." \
  --thought-number 4 \
  --total-thoughts 6 \
  --branch-from 2 \
  --branch-id "alt-approach"

# Final thought (complete chain)
python scripts/sequential_think_cli.py think \
  --thought "Conclusion: the solution is..." \
  --thought-number 5 \
  --total-thoughts 5 \
  --no-next

# View thought history
python scripts/sequential_think_cli.py history [--format json|text]

# Clear thought history
python scripts/sequential_think_cli.py clear

Core Principles

Iterative Thinking Process

Each tool call = one "thought" in the chain
Build upon, question, or revise previous thoughts
Express uncertainty when it exists

Dynamic Thought Count

Start with initial estimate of totalThoughts
Adjust up/down as understanding evolves
Add more thoughts even after reaching initial end

Hypothesis-Driven Approach

Generate hypotheses as potential solutions emerge
Verify hypotheses based on chain-of-thought steps
Repeat until satisfied with solution

Completion Criteria

Only set nextThoughtNeeded: false when truly finished
Must have satisfactory, verified answer
Don't rush to conclusion

When to Use

| Scenario | Use Sequential Think | |----------|---------------------| | Complex debugging (3+ layers) | ✅ Yes | | Architectural analysis | ✅ Yes | | Multi-component investigation | ✅ Yes | | Performance bottleneck analysis | ✅ Yes | | Root cause analysis | ✅ Yes | | Simple explanation | ❌ No | | Single-file change | ❌ No | | Straightforward fix | ❌ No |

Parameters

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | thought | string | Yes | Current thinking step content | | thoughtNumber | int | Yes | Current position in sequence (1-based) | | totalThoughts | int | Yes | Estimated total thoughts needed | | nextThoughtNeeded | bool | No | Whether more thinking needed (default: true) | | isRevision | bool | No | Whether this revises previous thinking | | revisesThought | int | No | Which thought number is being reconsidered | | branchFromThought | int | No | Branching point thought number | | branchId | string | No | Identifier for current branch | | needsMoreThoughts | bool | No | Signal that more thoughts needed beyond estimate |

Output Format

{
  "thoughtNumber": 3,
  "totalThoughts": 5,
  "nextThoughtNeeded": true,
  "branches": ["alt-approach"],
  "thoughtHistoryLength": 3
}

Workflow Pattern

Phase 1: Problem Decomposition

Thought 1: Identify problem scope and constraints
Thought 2: Break into sub-problems
Thought 3: Identify dependencies between sub-problems

Phase 2: Hypothesis Generation

Thought 4: Generate initial hypothesis
Thought 5: Identify evidence needed to verify

Phase 3: Verification & Iteration

Thought 6: Test hypothesis against evidence
Thought 7: Revise if needed (isRevision=true)
Thought 8: Branch if alternative path promising

Phase 4: Conclusion

Final Thought: Synthesize findings, provide answer (nextThoughtNeeded=false)

Best Practices

Start with estimate, adjust as needed
- Initial totalThoughts is just a guess
- Increase if problem more complex than expected
- Decrease if solution found early
Use revisions for course correction
- Mark isRevision=true when reconsidering
- Reference revisesThought for clarity
Branch for alternative approaches
- Use branchFromThought to explore alternatives
- Give meaningful branchId names
Filter irrelevant information
- Each thought should advance toward solution
- Ignore tangential details
Don't rush completion
- Only nextThoughtNeeded=false when truly done
- Verify hypothesis before concluding

Anti-Patterns

| Prohibited | Correct | |------------|---------| | Use for simple tasks | Reserve for complex multi-step problems | | Skip thought numbers | Always increment correctly | | Conclude without verification | Verify hypothesis before final thought | | Ignore previous thoughts | Build upon or explicitly revise | | Fixed totalThoughts | Adjust as understanding evolves |

Integration with Other Tools

With ACE-Tool

ACE-Tool → align current state → Sequential Think → analyze and plan

With Context7

Sequential Think → coordinate analysis → Context7 → provide official patterns

With Serena

Serena → symbol-level exploration → Sequential Think → systematic analysis

Agent Skills: Sequential Think

Install this agent skill to your local

Skill Files