Deep Reasoning

Deep Reasoning Skill

Use this skill when you need to think through complex problems systematically. The deep_reasoning tool helps you:

• Break down complex problems into manageable steps
• Explore multiple solution approaches (branching)
• Revise earlier thinking when new information emerges
• Generate and test hypotheses
• Track confidence levels in your conclusions
• Maintain structured reasoning across long analyses

When to Use Deep Reasoning

Use this tool when:

• The problem requires multiple steps to solve
• You need to explore alternative approaches
• Earlier assumptions might need revision
• You want to track confidence in your conclusions
• The problem involves hypothesis testing
• You need to maintain reasoning context across many steps

Don't use for:

• Simple factual lookups
• Single-step operations
• Tasks that don't benefit from structured thinking

The deep_reasoning Tool

Basic Parameters

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | thought | string | Yes | Your current thinking step | | thought_number | int | Yes | Current step number (1, 2, 3...) | | total_thoughts | int | Yes | Estimated total steps needed | | next_thought_needed | bool | Yes | True if more thinking needed |

Advanced Parameters

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | thought_type | string | "regular" | Type: regular, hypothesis, verification, revision, branch, conclusion, question, evidence | | reasoning_mode | string | "analytical" | Mode: analytical, creative, critical, exploratory, deductive, inductive, abductive | | confidence | float | 0.7 | Confidence level (0.0 to 1.0) | | is_revision | bool | false | Whether this revises a previous thought | | revises_thought | int | null | Which thought number is being revised | | branch_from_thought | int | null | Create a branch from this thought | | branch_id | string | null | Identifier for the branch | | hypothesis_statement | string | null | Propose a new hypothesis | | hypothesis_id | string | null | Reference an existing hypothesis |

Reasoning Modes

Choose the mode that fits your analysis:

| Mode | When to Use | |------|-------------| | analytical | Breaking down problems, examining components | | creative | Brainstorming, generating novel solutions | | critical | Evaluating claims, finding weaknesses | | exploratory | Open-ended investigation, discovery | | deductive | Applying general rules to specific cases | | inductive | Inferring patterns from specific examples | | abductive | Finding the best explanation for observations |

Thought Types

Mark your thoughts appropriately:

| Type | Purpose | |------|---------| | regular | Standard reasoning step | | hypothesis | Proposing something to test | | verification | Testing a hypothesis | | revision | Updating earlier thinking | | branch | Exploring an alternative path | | conclusion | Final determination | | question | Identifying what needs answering | | evidence | Recording supporting data |

Example Workflows

Basic Sequential Reasoning

Thought 1/5: "Let me understand the problem first..."
Thought 2/5: "The key constraints are..."
Thought 3/5: "One approach would be..."
Thought 4/5: "Evaluating this approach..."
Thought 5/5: "My conclusion is..." (next_thought_needed=false)

With Branching

Thought 1/5: "Initial analysis..."
Thought 2/5: "Approach A would..."
Thought 3/5: "But what if we try Approach B?" (branch_from_thought=2, branch_id="approach_b")
Thought 4/5: "Comparing both approaches..."
Thought 5/5: "Approach B is better because..."

With Hypothesis Testing

Thought 1/5: "Observing the symptoms..."
Thought 2/5: "I hypothesize the cause is X" (hypothesis_statement="The bug is in the auth layer")
Thought 3/5: "Testing this hypothesis..." (thought_type="verification", hypothesis_id="...")
Thought 4/5: "Evidence supports/refutes the hypothesis..."
Thought 5/5: "Conclusion based on verified hypothesis..."

With Revision

Thought 1/5: "Initial assumption: the database is the bottleneck"
Thought 2/5: "But wait, the metrics show..." (is_revision=true, revises_thought=1)
Thought 3/5: "Actually, the network latency is the issue..."

Complete Tool Call Examples

Example 1: Starting a Reasoning Session

{
  "thought": "Let me break down this authentication bug. The user reports they can log in but get logged out after 5 minutes.",
  "thought_number": 1,
  "total_thoughts": 6,
  "next_thought_needed": true,
  "reasoning_mode": "analytical",
  "confidence": 0.8
}

Example 2: Proposing a Hypothesis

{
  "thought": "Based on the 5-minute timeout, I suspect the session token expiration is set incorrectly.",
  "thought_number": 2,
  "total_thoughts": 6,
  "next_thought_needed": true,
  "thought_type": "hypothesis",
  "hypothesis_statement": "Session token TTL is set to 5 minutes instead of the expected 30 minutes",
  "confidence": 0.7
}

Example 3: Creating a Branch

{
  "thought": "Alternatively, this could be a cookie domain mismatch issue causing the session to not persist.",
  "thought_number": 3,
  "total_thoughts": 6,
  "next_thought_needed": true,
  "branch_from_thought": 2,
  "branch_id": "cookie_theory",
  "reasoning_mode": "exploratory",
  "confidence": 0.5
}

Example 4: Revising Earlier Thinking

{
  "thought": "After checking the config, the TTL is correct at 30 minutes. My earlier hypothesis was wrong.",
  "thought_number": 4,
  "total_thoughts": 6,
  "next_thought_needed": true,
  "is_revision": true,
  "revises_thought": 2,
  "thought_type": "revision",
  "confidence": 0.9
}

Example 5: Reaching a Conclusion

{
  "thought": "The issue is confirmed: the cookie domain was set to 'example.com' but the app runs on 'app.example.com'. Fixing the domain setting resolves the logout issue.",
  "thought_number": 6,
  "total_thoughts": 6,
  "next_thought_needed": false,
  "thought_type": "conclusion",
  "confidence": 0.95
}

Best Practices

1. Start with Clear Problem Definition

Always begin by clearly stating what you're trying to solve. This anchors the entire reasoning chain.

2. Adjust Total Thoughts Dynamically

If you realize you need more steps, increase total_thoughts. The tool handles this gracefully.

3. Use Appropriate Confidence Levels

• 0.9-1.0: Very confident, based on verified facts
• 0.7-0.8: Reasonably confident, good evidence
• 0.5-0.6: Uncertain, exploring possibilities
• 0.3-0.4: Low confidence, speculative
• 0.1-0.2: Highly uncertain, just brainstorming

4. Branch When Exploring Alternatives

Don't just mention alternatives—create explicit branches. This helps track which path led to the solution.

5. Revise Rather Than Ignore

When you realize earlier thinking was wrong, use is_revision=true to explicitly update it. This maintains reasoning integrity.

6. End with Conclusions

Always set next_thought_needed=false and thought_type="conclusion" when you've reached your answer.

Tool Response Format

The tool returns a JSON response with:

{
  "status": "success",
  "thought_id": "abc123",
  "thought_number": 3,
  "total_thoughts": 6,
  "next_thought_needed": true,
  "branches": ["main", "cookie_theory"],
  "hypotheses": ["hyp_001"],
  "thought_history_length": 3,
  "average_confidence": 0.73,
  "current_mode": "analytical",
  "guidance": "Consider testing your hypothesis with evidence.",
  "hypothesis_id": "hyp_001"
}

Integration with Other Tools

Deep reasoning works well with:

• web_search: Gather evidence to support or refute hypotheses
• read_file: Examine code or configs during analysis
• run_skill: Execute other skills as part of your reasoning
• memory_search: Recall relevant past context

Troubleshooting

"Max branches reached"

You've hit the branch limit (default: 10). Consolidate branches or start a new reasoning session.

"Max thoughts reached"

You've hit the thought limit (default: 50). Summarize your findings and start fresh if needed.

Low average confidence

If your average confidence is below 0.5, consider:

• Gathering more evidence
• Revising uncertain assumptions
• Exploring alternative approaches

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: markdevshop
• Source: markdevshop/AG3NT
• License: MIT

Install and usage instructions live in the source repository linked above.