🧠 Sequential Thinking

The Sequential Thinking MCP enables systematic, verifiable, step-by-step reasoning via an agentic plan-execute-reflect loop, allowing the AI to dynamically correct course.

Domain: wiki.omp.loca.zone | ID: omp-sequential-thinking | Version: 1.0.0

Native OMP extension. Enforces cognitive steps before tool execution via ExtensionAPI. Zero-latency. In-memory state. Active blocking.

---

🔌 Architecture Hooks

Replaces stdio MCP:

• pi.registerFlag: CLI args
• pi.getFlag: CLI args
• pi.registerCommand: TUI slash commands
• pi.registerTool: Native sequential_thinking tool
• pi.on(context): Injects forcedThoughts instruction
• pi.on(tool_call): Blocks non-cognitive tools until threshold met
• pi.on(message_end): Loops prompt if LLM yields early
• pi.on(turn_start): Counter lifecycle
• pi.on(tool_execution_end): Counter lifecycle

📦 Layout Deployment

cd ~/.omp/plugins
bun add file:../extensions/sequential-thinking

Manifest:

• File: package.json
• Type: module
• Entry: ./src/index.ts via omp.extensions

---

🗃️ Parameters / Tool Schema

Parameter	Type	Description
thought	string	The actual reasoning text for the current step.
thoughtNumber	integer	The current step number in the thinking sequence.
totalThoughts	integer	The total estimated thoughts required (dynamic).
nextThoughtNeeded	boolean	Set to true to continue the loop, or false to output the final response.
isRevision	boolean	Indicates if this step corrects a previous thought.
revisesThought	integer	The specific thought number being corrected.
branchFromThought	integer	The thought number from which to fork a new reasoning branch.
branchId	string	Unique identifier for the active reasoning branch.
needsMoreThoughts	boolean	Request to increase totalThoughts when limits are reached.

Example Payload:

{
  "thought": "The database migration failed because of a missing index on the 'users' table. I need to create a branch to test the index creation before altering the production DB.",
  "thoughtNumber": 4,
  "totalThoughts": 6,
  "nextThoughtNeeded": true,
  "branchFromThought": 3,
  "branchId": "test-index-creation"
}

---

🔁 Execution Loop

1. Plan: Initialize sequential_thinking (thoughtNumber: 1, nextThoughtNeeded: true) to define the execution plan.
2. Execute: Interleave external tool calls (e.g., read, bash).
3. Reflect: Call sequential_thinking to process tool output.
4. Terminate: Call sequential_thinking (nextThoughtNeeded: false).
5. Synthesize: Yield final Markdown output.

---

🎯 Best Practices & Rules

1. Interleave Actions: Alternate between thinking steps and tool executions (e.g., Think $\to$ Read File $\to$ Think $\to$ Modify Code).
2. Structure Phases: Follow sequence: Define $\to$ Research $\to$ Analyze $\to$ Synthesize.
3. Atomic Steps: Limit to one logical leap per thought. No monolithic dumps.
4. Pivot Safely: Use isRevision to fix flawed premises immediately. Use branchFromThought to test alternative strategies. Never hallucinate forward.
5. State Tracking: Define the branch goal in the first thought of a new branch.

Common Pitfalls

• Monolithic Thoughts: Packing multiple logical phases into a single step. Keep thoughts atomic.
• Infinite Loops: Setting nextThoughtNeeded: true repeatedly without progression. Revise logic if stuck.
• Undocumented Branches: Forking paths without clarifying the branch goal in the initial thought of that branch.

---

🛡️ Enforcement Sandbox & Controls

• CLI: omp --seq-thoughts <N> "<prompt>"

• TUI: /seqthink <N>

• Context: Injects developer role forcing tool usage.

• Blocker: Rejects premature tool calls with block reason.

• Loop: Sends followUp if agent skips thinking.

---

🚦 When to Use vs. Skip

Use Case	Recommended?	Why
Complex Architecture / DB Migrations	Yes	Requires multi-stage planning and verification.
Deep Debugging / Memory Leaks	Yes	Requires systematic hypothesis testing.
Ambiguous Tasks / Refactoring	Yes	Requires iterative exploration.
Simple Math / Facts	No	Unnecessary overhead.
Low-Latency Queries	No	Sequential iterations add execution time.
Single-Step Tasks (Summarization)	No	No reasoning depth needed.

🧪 QA Validation

1. Deps: Verify omp-sequential-thinking in plugin deps.
2. CLI: Run omp --help to confirm --seq-thoughts.
3. Loop: Run omp --seq-thoughts 3 "Calc." and verify 3 sequential thoughts.
4. Block: Run /seqthink 5 then “Run ls” and verify bash block.

---