Reasoning Tools¶

The ReasoningTools toolkit provides text-first reasoning utilities for AI agents:

• Build a compact reasoning chain (steps + reflections + scratchpad)
• Optionally detect simple bias markers in text
• Assess chain quality and suggest improvements
• Synthesize the chain into a short conclusion/summary/insight

All public functions return strings (human-readable, markdown-ish).

🤖 AI Agent Setup (Agno)¶

from agno.agent import Agent
from enhancedtoolkits import ReasoningTools

agent = Agent(
    name="Analyst",
    model="gpt-4",
    tools=[ReasoningTools(reasoning_depth=5, enable_bias_detection=True)],
)

The agent_or_team parameter should be your agent/team object (it is used as the session state container).

🔧 Key Concepts¶

Valid values¶

• cognitive_mode: analysis | synthesis | evaluation | planning | creative | reflection
• reasoning_type: deductive | inductive | abductive | causal | probabilistic | analogical
• synthesis_type: conclusion | summary | insights
• scratchpad operation: set | get | list | clear

🧠 Available Functions¶

add_structured_reasoning_step(agent_or_team, problem, cognitive_mode='analysis', reasoning_type='deductive', evidence=None, confidence=0.5)¶

Adds a numbered reasoning step and (optionally) records evidence and bias markers.

add_meta_cognitive_reflection(agent_or_team, reflection, step_id=None)¶

Adds a reflection entry for the current reasoning chain.

manage_working_memory_scratchpad(agent_or_team, key, value=None, operation='set')¶

Simple key/value scratchpad: - set: set key=value - get: retrieve one key - list: list all keys - clear: clear one key or all keys (key='all')

assess_reasoning_quality_and_suggest_improvements(agent_or_team)¶

Computes a compact quality score and suggests improvements.

synthesize_reasoning_chain_into_conclusion_or_insight(agent_or_team, synthesis_type='conclusion')¶

Produces a short synthesis and marks the chain as completed.

retrieve_current_reasoning_session_state(agent_or_team)¶

Returns a compact summary of the current chain state.

reset_reasoning_session_state(agent_or_team)¶

Clears the current reasoning session.

✅ Example Workflow¶

from enhancedtoolkits import ReasoningTools

reasoning = ReasoningTools()

# 1) Add a step
reasoning.add_structured_reasoning_step(
    agent_or_team=agent,
    problem="Should we invest in renewable energy stocks?",
    cognitive_mode="analysis",
    reasoning_type="inductive",
    evidence=["Costs are declining", "Policy incentives are rising"],
    confidence=0.6,
)

# 2) Add a reflection
reasoning.add_meta_cognitive_reflection(
    agent_or_team=agent,
    reflection="We may be overweighting recent headlines; check longer-term fundamentals.",
    step_id=1,
)

# 3) Store something in scratchpad
reasoning.manage_working_memory_scratchpad(
    agent_or_team=agent,
    key="tickers",
    value="ICLN,TAN",
    operation="set",
)

# 4) Assess quality
reasoning.assess_reasoning_quality_and_suggest_improvements(agent_or_team=agent)

# 5) Synthesize
reasoning.synthesize_reasoning_chain_into_conclusion_or_insight(
    agent_or_team=agent,
    synthesis_type="conclusion",
)

# 6) Reset
reasoning.reset_reasoning_session_state(agent_or_team=agent)

📝 Notes / Best Practices¶

• Keep each step short (1–3 sentences). Store bulky details in evidence or scratchpad keys.
• Prefer synthesis outputs instead of returning the entire chain verbatim.

API Reference¶

• docs/api/reasoning.md