Cognitive Policies¶
When making important decisions, humans naturally gather information first, mentally rehearse plans before acting, and assess whether information is reliable. Cognitive Policies bring this same multi-round deliberation to your agents — all within a single OTC (observe-think-act) cycle.
There are three cognitive policies:
- Acquiring — progressive disclosure of skill instructions and history details on demand
- Rehearsal — mental simulation of a planned action before committing
- Reflection — assessment of information quality and consistency
In this tutorial, we'll focus on Acquiring, which is the built-in, always-active policy and has the most practical use cases: skill detail disclosure and history detail disclosure. We'll briefly cover Rehearsal and Reflection at the end.
Initialize¶
In [1]:
Copied!
import os
model_name = os.environ.get("MODEL_NAME")
api_key = os.environ.get("API_KEY")
api_base = os.environ.get("BASE_URL")
import os model_name = os.environ.get("MODEL_NAME") api_key = os.environ.get("API_KEY") api_base = os.environ.get("BASE_URL")
In [ ]:
Copied!
from bridgic.llms.openai import OpenAILlm, OpenAIConfiguration
llm = OpenAILlm(
api_key=api_key,
api_base=api_base,
timeout=120,
configuration=OpenAIConfiguration(
model=model_name,
temperature=0.0,
max_tokens=16384,
),
)
from bridgic.llms.openai import OpenAILlm, OpenAIConfiguration llm = OpenAILlm( api_key=api_key, api_base=api_base, timeout=120, configuration=OpenAIConfiguration( model=model_name, temperature=0.0, max_tokens=16384, ), )
In [3]:
Copied!
from bridgic.core.agentic.tool_specs import FunctionToolSpec
async def get_stock_price(ticker: str) -> str:
"""Fetch stock price data for a ticker symbol"""
prices = {"AAPL": "$178.50", "GOOGL": "$141.20", "MSFT": "$415.30", "TSLA": "$245.80"}
return f"{ticker}: {prices.get(ticker, 'Unknown ticker')}"
async def get_financials(company: str) -> str:
"""Get financial metrics for a company"""
return f"{company} - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45"
async def analyze_sentiment(ticker: str) -> str:
"""Analyze market sentiment for a stock"""
return f"Sentiment for {ticker}: 65% bullish, 25% neutral, 10% bearish"
async def generate_recommendation(ticker: str, analysis: str) -> str:
"""Generate an investment recommendation report"""
return f"Investment Report for {ticker}: {analysis}"
get_stock_price_tool = FunctionToolSpec.from_raw(get_stock_price)
get_financials_tool = FunctionToolSpec.from_raw(get_financials)
analyze_sentiment_tool = FunctionToolSpec.from_raw(analyze_sentiment)
generate_recommendation_tool = FunctionToolSpec.from_raw(generate_recommendation)
from bridgic.core.agentic.tool_specs import FunctionToolSpec async def get_stock_price(ticker: str) -> str: """Fetch stock price data for a ticker symbol""" prices = {"AAPL": "$178.50", "GOOGL": "$141.20", "MSFT": "$415.30", "TSLA": "$245.80"} return f"{ticker}: {prices.get(ticker, 'Unknown ticker')}" async def get_financials(company: str) -> str: """Get financial metrics for a company""" return f"{company} - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45" async def analyze_sentiment(ticker: str) -> str: """Analyze market sentiment for a stock""" return f"Sentiment for {ticker}: 65% bullish, 25% neutral, 10% bearish" async def generate_recommendation(ticker: str, analysis: str) -> str: """Generate an investment recommendation report""" return f"Investment Report for {ticker}: {analysis}" get_stock_price_tool = FunctionToolSpec.from_raw(get_stock_price) get_financials_tool = FunctionToolSpec.from_raw(get_financials) analyze_sentiment_tool = FunctionToolSpec.from_raw(analyze_sentiment) generate_recommendation_tool = FunctionToolSpec.from_raw(generate_recommendation)
Acquiring — Progressive Disclosure on Demand¶
Acquiring is the built-in, always-active cognitive policy. It enables progressive disclosure through LayeredExposure fields — specifically skills and cognitive_history.
Here's how it works:
- The LLM sees summaries of skills and history steps (not the full content)
- If the LLM needs more detail, it fills the
detailsfield in its output with a list of items to reveal - The framework reveals the requested full content and re-triggers thinking
- In the next round, the LLM sees the disclosed details and can make an informed decision
Acquiring fires at most once per OTC cycle, then closes. It is always active — no flag needed to enable it. It is only disabled when output_schema is set on the worker.
What the LLM sees across rounds:
- Round 1 output instructions include a
detailsfield:List[{field: str, index: int}] - After the LLM requests details, Round 2 gets the disclosed details injected into the context and the
detailsfield is removed from the output instructions
Acquiring Flow¶
Round 1: LLM sees skill summaries → fills details: [{field: "skills", index: 0}]
↓
Framework: reveals skill[0] full content, removes details field from output schema
↓
Round 2: LLM sees disclosed details → makes tool call decision
Part 1: Acquiring Skill Details¶
Skills use LayeredExposure — the LLM initially sees only the skill name and description. If a skill matches its task, the LLM can request the full instructions via the details field.
This is perfect for giving agents access to many procedures without flooding the context window. The agent only loads the procedure it actually needs.
In [4]:
Copied!
from bridgic.amphibious import Skill
fundamental_skill = Skill(
name="fundamental-analysis",
description="Evaluate a stock's intrinsic value using financial metrics and ratios",
content=(
"## Fundamental Analysis Procedure\n\n"
"Follow these steps to evaluate a stock:\n\n"
"1. **Get Financial Data**: Retrieve the company's key metrics using get_financials.\n"
"2. **Evaluate P/E Ratio**:\n"
" - P/E < 15: Undervalued (positive signal)\n"
" - P/E 15-25: Fairly valued\n"
" - P/E > 25: Potentially overvalued (caution)\n"
"3. **Assess Growth**: Revenue growth > 10% is strong, > 20% is exceptional.\n"
"4. **Check Debt**: Debt/Equity < 0.5 is healthy, > 1.0 is concerning.\n"
"5. **Get Current Price**: Use get_stock_price for the latest quote.\n"
"6. **Rating Logic**:\n"
" - BUY: P/E < 20 AND growth > 10% AND debt/equity < 0.7\n"
" - SELL: P/E > 35 OR growth < 0% OR debt/equity > 1.5\n"
" - HOLD: Everything else\n"
"7. **Generate Report**: Use generate_recommendation with your analysis summary."
),
)
technical_skill = Skill(
name="technical-analysis",
description="Analyze stock price patterns, momentum, and technical indicators",
content=(
"## Technical Analysis Procedure\n\n"
"1. Get current and historical price data.\n"
"2. Calculate RSI (Relative Strength Index):\n"
" - RSI > 70: Overbought (potential sell signal)\n"
" - RSI < 30: Oversold (potential buy signal)\n"
"3. Identify support and resistance levels from price history.\n"
"4. Check momentum: compare recent price to 50-day moving average.\n"
"5. Generate a technical outlook (bullish/bearish/neutral)."
),
)
risk_skill = Skill(
name="portfolio-risk-assessment",
description="Evaluate portfolio risk using volatility, sentiment, and diversification metrics",
content=(
"## Portfolio Risk Assessment Procedure\n\n"
"1. Gather sentiment data for each holding using analyze_sentiment.\n"
"2. Calculate overall portfolio sentiment score.\n"
"3. Assess concentration risk: single position > 20% is high risk.\n"
"4. Evaluate volatility based on price fluctuations.\n"
"5. Assign risk rating: LOW / MEDIUM / HIGH.\n"
"6. Recommend rebalancing if any position exceeds thresholds."
),
)
from bridgic.amphibious import Skill fundamental_skill = Skill( name="fundamental-analysis", description="Evaluate a stock's intrinsic value using financial metrics and ratios", content=( "## Fundamental Analysis Procedure\n\n" "Follow these steps to evaluate a stock:\n\n" "1. **Get Financial Data**: Retrieve the company's key metrics using get_financials.\n" "2. **Evaluate P/E Ratio**:\n" " - P/E < 15: Undervalued (positive signal)\n" " - P/E 15-25: Fairly valued\n" " - P/E > 25: Potentially overvalued (caution)\n" "3. **Assess Growth**: Revenue growth > 10% is strong, > 20% is exceptional.\n" "4. **Check Debt**: Debt/Equity < 0.5 is healthy, > 1.0 is concerning.\n" "5. **Get Current Price**: Use get_stock_price for the latest quote.\n" "6. **Rating Logic**:\n" " - BUY: P/E < 20 AND growth > 10% AND debt/equity < 0.7\n" " - SELL: P/E > 35 OR growth < 0% OR debt/equity > 1.5\n" " - HOLD: Everything else\n" "7. **Generate Report**: Use generate_recommendation with your analysis summary." ), ) technical_skill = Skill( name="technical-analysis", description="Analyze stock price patterns, momentum, and technical indicators", content=( "## Technical Analysis Procedure\n\n" "1. Get current and historical price data.\n" "2. Calculate RSI (Relative Strength Index):\n" " - RSI > 70: Overbought (potential sell signal)\n" " - RSI < 30: Oversold (potential buy signal)\n" "3. Identify support and resistance levels from price history.\n" "4. Check momentum: compare recent price to 50-day moving average.\n" "5. Generate a technical outlook (bullish/bearish/neutral)." ), ) risk_skill = Skill( name="portfolio-risk-assessment", description="Evaluate portfolio risk using volatility, sentiment, and diversification metrics", content=( "## Portfolio Risk Assessment Procedure\n\n" "1. Gather sentiment data for each holding using analyze_sentiment.\n" "2. Calculate overall portfolio sentiment score.\n" "3. Assess concentration risk: single position > 20% is high risk.\n" "4. Evaluate volatility based on price fluctuations.\n" "5. Assign risk rating: LOW / MEDIUM / HIGH.\n" "6. Recommend rebalancing if any position exceeds thresholds." ), )
In [ ]:
Copied!
from bridgic.amphibious import (
AmphibiousAutoma, CognitiveContext, CognitiveWorker, think_unit
)
class InvestmentAnalyst(AmphibiousAutoma[CognitiveContext]):
analyst = think_unit(
CognitiveWorker.inline(
"You are an investment analyst. Follow the appropriate skill procedure. "
"If a skill is available that matches your task, request its details first.",
verbose_prompt=True,
),
max_attempts=6,
)
async def on_agent(self, ctx: CognitiveContext):
await self.analyst
agent = InvestmentAnalyst(llm=llm, verbose=True)
result = await agent.arun(
goal="Perform a fundamental analysis of AAPL and give a BUY/HOLD/SELL rating",
tools=[get_stock_price_tool, get_financials_tool, analyze_sentiment_tool, generate_recommendation_tool],
skills=[fundamental_skill, technical_skill, risk_skill],
)
print(result)
from bridgic.amphibious import ( AmphibiousAutoma, CognitiveContext, CognitiveWorker, think_unit ) class InvestmentAnalyst(AmphibiousAutoma[CognitiveContext]): analyst = think_unit( CognitiveWorker.inline( "You are an investment analyst. Follow the appropriate skill procedure. " "If a skill is available that matches your task, request its details first.", verbose_prompt=True, ), max_attempts=6, ) async def on_agent(self, ctx: CognitiveContext): await self.analyst agent = InvestmentAnalyst(llm=llm, verbose=True) result = await agent.arun( goal="Perform a fundamental analysis of AAPL and give a BUY/HOLD/SELL rating", tools=[get_stock_price_tool, get_financials_tool, analyze_sentiment_tool, generate_recommendation_tool], skills=[fundamental_skill, technical_skill, risk_skill], ) print(result)
[18:03:40.348] [Router] (_amphibious_automa.py:1573) Auto-detecting execution mode [18:03:40.348] [Router] (_amphibious_automa.py:1579) Detected AGENT mode [18:03:40.349] [Observe] (_amphibious_automa.py:861) _PromptWorker: None [18:03:45.256] [Think] (_cognitive_worker.py:332) Message 1 (system, 476 tokens): You are an investment analyst. Follow the appropriate skill procedure. If a skill is available that matches your task, request its details first.Respond in JSON format. # Available Tools (with parameters): • get_stock_price: Fetch stock price data for a ticker symbol - ticker (string) [required] • get_financials: Get financial metrics for a company - company (string) [required] • analyze_sentiment: Analyze market sentiment for a stock - ticker (string) [required] • generate_recommendation: Generate an investment recommendation report - ticker (string) [required] - analysis (string) [required] # Available Skills (request details via details: {field: 'skills', index: N}): [0] fundamental-analysis - Evaluate a stock's intrinsic value using financial metrics and ratios [1] technical-analysis - Analyze stock price patterns, momentum, and technical indicators [2] portfolio-risk-assessment - Evaluate portfolio risk using volatility, sentiment, and diversification metrics # Context Acquiring If the context contains progressively disclosed information (e.g. skills, history steps) and you want to inspect the details, use the **details** field to request them. The framework will expand these items in the next round. Batch all requests in a single output. When using this field, leave step_content and output empty. ## Field format: - **details**: [{field: "skills", index: 0}, ...] Available fields: **skills** (view a skill's full workflow), **cognitive_history** (view the full result of a previous step) # Output Fields - **step_content**: Your analysis and reasoning for this step - **finish**: Set True when the sub-task is fully complete (default: False) - **details**: Available fields: **skills**, **cognitive_history**. example: [{field: 'skills', index: 0}, ...] - **output**: Tool calls to execute: [{tool, tool_arguments: [{name: 'param', value: 'value'}]}] [18:03:45.256] [Think] (_cognitive_worker.py:332) Message 2 (user, 39 tokens): Based on the context below, decide your next action. Goal: Perform a fundamental analysis of AAPL and give a BUY/HOLD/SELL rating Execution History: (none) [18:03:45.256] [Think] (_cognitive_worker.py:332) Total: 515 tokens (cumulative: 515) [18:03:53.539] [Think] (_cognitive_worker.py:332) Message 1 (system, 313 tokens): You are an investment analyst. Follow the appropriate skill procedure. If a skill is available that matches your task, request its details first.Respond in JSON format. # Available Tools (with parameters): • get_stock_price: Fetch stock price data for a ticker symbol - ticker (string) [required] • get_financials: Get financial metrics for a company - company (string) [required] • analyze_sentiment: Analyze market sentiment for a stock - ticker (string) [required] • generate_recommendation: Generate an investment recommendation report - ticker (string) [required] - analysis (string) [required] # Available Skills (request details via details: {field: 'skills', index: N}): [0] fundamental-analysis - Evaluate a stock's intrinsic value using financial metrics and ratios [1] technical-analysis - Analyze stock price patterns, momentum, and technical indicators [2] portfolio-risk-assessment - Evaluate portfolio risk using volatility, sentiment, and diversification metrics # Output Fields - **step_content**: Your analysis and reasoning for this step - **finish**: Set True when the sub-task is fully complete (default: False) - **output**: Tool calls to execute: [{tool, tool_arguments: [{name: 'param', value: 'value'}]}] [18:03:53.539] [Think] (_cognitive_worker.py:332) Message 2 (user, 284 tokens): Based on the context below, decide your next action. Goal: Perform a fundamental analysis of AAPL and give a BUY/HOLD/SELL rating Execution History: (none) Previously Disclosed Details: [skills[0]]: Skill: fundamental-analysis Description: Evaluate a stock's intrinsic value using financial metrics and ratios Instructions: ---------------------------------------- ## Fundamental Analysis Procedure Follow these steps to evaluate a stock: 1. **Get Financial Data**: Retrieve the company's key metrics using get_financials. 2. **Evaluate P/E Ratio**: - P/E < 15: Undervalued (positive signal) - P/E 15-25: Fairly valued - P/E > 25: Potentially overvalued (caution) 3. **Assess Growth**: Revenue growth > 10% is strong, > 20% is exceptional. 4. **Check Debt**: Debt/Equity < 0.5 is healthy, > 1.0 is concerning. 5. **Get Current Price**: Use get_stock_price for the latest quote. 6. **Rating Logic**: - BUY: P/E < 20 AND growth > 10% AND debt/equity < 0.7 - SELL: P/E > 35 OR growth < 0% OR debt/equity > 1.5 - HOLD: Everything else 7. **Generate Report**: Use generate_recommendation with your analysis summary. [18:03:53.539] [Think] (_cognitive_worker.py:332) Total: 597 tokens (cumulative: 1112) [18:03:53.541] [Think] (_amphibious_automa.py:867) _PromptWorker: finish=False, step=Starting fundamental analysis for AAPL. First step is to retrieve financial metrics to evaluate P/E ratio, revenue growth, and debt/equity. [18:03:53.544] [Act] (_amphibious_automa.py:873) _PromptWorker: { "content": "Starting fundamental analysis for AAPL. First step is to retrieve financial metrics to evaluate P/E ratio, revenue growth, and debt/equity.", "result": { "results": [ { "tool_id": "call_0", "tool_name": "get_financials", "tool_arguments": { "company": "AAPL" }, "tool_result": "AAPL - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45", "success": true, "error": null } ] }, "metadata": {}, "status": null } [18:03:53.545] [Observe] (_amphibious_automa.py:861) _PromptWorker: None [18:04:28.142] [Think] (_cognitive_worker.py:332) Message 1 (system, 476 tokens): You are an investment analyst. Follow the appropriate skill procedure. If a skill is available that matches your task, request its details first.Respond in JSON format. # Available Tools (with parameters): • get_stock_price: Fetch stock price data for a ticker symbol - ticker (string) [required] • get_financials: Get financial metrics for a company - company (string) [required] • analyze_sentiment: Analyze market sentiment for a stock - ticker (string) [required] • generate_recommendation: Generate an investment recommendation report - ticker (string) [required] - analysis (string) [required] # Available Skills (request details via details: {field: 'skills', index: N}): [0] fundamental-analysis - Evaluate a stock's intrinsic value using financial metrics and ratios [1] technical-analysis - Analyze stock price patterns, momentum, and technical indicators [2] portfolio-risk-assessment - Evaluate portfolio risk using volatility, sentiment, and diversification metrics # Context Acquiring If the context contains progressively disclosed information (e.g. skills, history steps) and you want to inspect the details, use the **details** field to request them. The framework will expand these items in the next round. Batch all requests in a single output. When using this field, leave step_content and output empty. ## Field format: - **details**: [{field: "skills", index: 0}, ...] Available fields: **skills** (view a skill's full workflow), **cognitive_history** (view the full result of a previous step) # Output Fields - **step_content**: Your analysis and reasoning for this step - **finish**: Set True when the sub-task is fully complete (default: False) - **details**: Available fields: **skills**, **cognitive_history**. example: [{field: 'skills', index: 0}, ...] - **output**: Tool calls to execute: [{tool, tool_arguments: [{name: 'param', value: 'value'}]}] [18:04:28.142] [Think] (_cognitive_worker.py:332) Message 2 (user, 380 tokens): Based on the context below, decide your next action. Goal: Perform a fundamental analysis of AAPL and give a BUY/HOLD/SELL rating Execution History: [Working Memory (0-0)] [0] Starting fundamental analysis for AAPL. First step is to retrieve financial metrics to evaluate P/E ratio, revenue growth, and debt/equity. Result: results=[ActionStepResult(tool_id='call_0', tool_name='get_financials', tool_arguments={'company': 'AAPL'}, tool_result='AAPL - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45', success=True, error=None)] Previously Disclosed Details: [skills[0]]: Skill: fundamental-analysis Description: Evaluate a stock's intrinsic value using financial metrics and ratios Instructions: ---------------------------------------- ## Fundamental Analysis Procedure Follow these steps to evaluate a stock: 1. **Get Financial Data**: Retrieve the company's key metrics using get_financials. 2. **Evaluate P/E Ratio**: - P/E < 15: Undervalued (positive signal) - P/E 15-25: Fairly valued - P/E > 25: Potentially overvalued (caution) 3. **Assess Growth**: Revenue growth > 10% is strong, > 20% is exceptional. 4. **Check Debt**: Debt/Equity < 0.5 is healthy, > 1.0 is concerning. 5. **Get Current Price**: Use get_stock_price for the latest quote. 6. **Rating Logic**: - BUY: P/E < 20 AND growth > 10% AND debt/equity < 0.7 - SELL: P/E > 35 OR growth < 0% OR debt/equity > 1.5 - HOLD: Everything else 7. **Generate Report**: Use generate_recommendation with your analysis summary. [18:04:28.142] [Think] (_cognitive_worker.py:332) Total: 856 tokens (cumulative: 1968) [18:04:28.144] [Think] (_amphibious_automa.py:867) _PromptWorker: finish=False, step=Retrieving current stock price for AAPL to complete fundamental analysis steps as per procedure. [18:04:28.147] [Act] (_amphibious_automa.py:873) _PromptWorker: { "content": "Retrieving current stock price for AAPL to complete fundamental analysis steps as per procedure.", "result": { "results": [ { "tool_id": "call_0", "tool_name": "get_stock_price", "tool_arguments": { "ticker": "AAPL" }, "tool_result": "AAPL: $178.50", "success": true, "error": null } ] }, "metadata": {}, "status": null } [18:04:28.147] [Observe] (_amphibious_automa.py:861) _PromptWorker: None [18:04:43.761] [Think] (_cognitive_worker.py:332) Message 1 (system, 476 tokens): You are an investment analyst. Follow the appropriate skill procedure. If a skill is available that matches your task, request its details first.Respond in JSON format. # Available Tools (with parameters): • get_stock_price: Fetch stock price data for a ticker symbol - ticker (string) [required] • get_financials: Get financial metrics for a company - company (string) [required] • analyze_sentiment: Analyze market sentiment for a stock - ticker (string) [required] • generate_recommendation: Generate an investment recommendation report - ticker (string) [required] - analysis (string) [required] # Available Skills (request details via details: {field: 'skills', index: N}): [0] fundamental-analysis - Evaluate a stock's intrinsic value using financial metrics and ratios [1] technical-analysis - Analyze stock price patterns, momentum, and technical indicators [2] portfolio-risk-assessment - Evaluate portfolio risk using volatility, sentiment, and diversification metrics # Context Acquiring If the context contains progressively disclosed information (e.g. skills, history steps) and you want to inspect the details, use the **details** field to request them. The framework will expand these items in the next round. Batch all requests in a single output. When using this field, leave step_content and output empty. ## Field format: - **details**: [{field: "skills", index: 0}, ...] Available fields: **skills** (view a skill's full workflow), **cognitive_history** (view the full result of a previous step) # Output Fields - **step_content**: Your analysis and reasoning for this step - **finish**: Set True when the sub-task is fully complete (default: False) - **details**: Available fields: **skills**, **cognitive_history**. example: [{field: 'skills', index: 0}, ...] - **output**: Tool calls to execute: [{tool, tool_arguments: [{name: 'param', value: 'value'}]}] [18:04:43.761] [Think] (_cognitive_worker.py:332) Message 2 (user, 450 tokens): Based on the context below, decide your next action. Goal: Perform a fundamental analysis of AAPL and give a BUY/HOLD/SELL rating Execution History: [Working Memory (0-1)] [0] Starting fundamental analysis for AAPL. First step is to retrieve financial metrics to evaluate P/E ratio, revenue growth, and debt/equity. Result: results=[ActionStepResult(tool_id='call_0', tool_name='get_financials', tool_arguments={'company': 'AAPL'}, tool_result='AAPL - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45', success=True, error=None)] [1] Retrieving current stock price for AAPL to complete fundamental analysis steps as per procedure. Result: results=[ActionStepResult(tool_id='call_0', tool_name='get_stock_price', tool_arguments={'ticker': 'AAPL'}, tool_result='AAPL: $178.50', success=True, error=None)] Previously Disclosed Details: [skills[0]]: Skill: fundamental-analysis Description: Evaluate a stock's intrinsic value using financial metrics and ratios Instructions: ---------------------------------------- ## Fundamental Analysis Procedure Follow these steps to evaluate a stock: 1. **Get Financial Data**: Retrieve the company's key metrics using get_financials. 2. **Evaluate P/E Ratio**: - P/E < 15: Undervalued (positive signal) - P/E 15-25: Fairly valued - P/E > 25: Potentially overvalued (caution) 3. **Assess Growth**: Revenue growth > 10% is strong, > 20% is exceptional. 4. **Check Debt**: Debt/Equity < 0.5 is healthy, > 1.0 is concerning. 5. **Get Current Price**: Use get_stock_price for the latest quote. 6. **Rating Logic**: - BUY: P/E < 20 AND growth > 10% AND debt/equity < 0.7 - SELL: P/E > 35 OR growth < 0% OR debt/equity > 1.5 - HOLD: Everything else 7. **Generate Report**: Use generate_recommendation with your analysis summary. [18:04:43.761] [Think] (_cognitive_worker.py:332) Total: 926 tokens (cumulative: 2894) [18:04:43.764] [Think] (_amphibious_automa.py:867) _PromptWorker: finish=False, step=Based on the financial metrics: P/E ratio of 28.5 (potentially overvalued), revenue growth of 12% (strong), and debt/equity of 0.45 (healthy). Applying rating logic: P/E >25 but <35, growth >10%, debt <0.7 → HOLD rating. Now generating the recommendation report. [18:04:43.766] [Act] (_amphibious_automa.py:873) _PromptWorker: { "content": "Based on the financial metrics: P/E ratio of 28.5 (potentially overvalued), revenue growth of 12% (strong), and debt/equity of 0.45 (healthy). Applying rating logic: P/E >25 but <35, growth >10%, debt <0.7 → HOLD rating. Now generating the recommendation report.", "result": { "results": [ { "tool_id": "call_0", "tool_name": "generate_recommendation", "tool_arguments": { "ticker": "AAPL", "analysis": "AAPL shows strong revenue growth (12%) and healthy debt levels (0.45 D/E), but P/E ratio of 28.5 suggests potential overvaluation. Meets HOLD criteria per fundamental analysis procedure." }, "tool_result": "Investment Report for AAPL: AAPL shows strong revenue growth (12%) and healthy debt levels (0.45 D/E), but P/E ratio of 28.5 suggests potential overvaluation. Meets HOLD criteria per fundamental analysis procedure.", "success": true, "error": null } ] }, "metadata": {}, "status": null } [18:04:43.767] [Observe] (_amphibious_automa.py:861) _PromptWorker: None [18:04:51.260] [Think] (_cognitive_worker.py:332) Message 1 (system, 476 tokens): You are an investment analyst. Follow the appropriate skill procedure. If a skill is available that matches your task, request its details first.Respond in JSON format. # Available Tools (with parameters): • get_stock_price: Fetch stock price data for a ticker symbol - ticker (string) [required] • get_financials: Get financial metrics for a company - company (string) [required] • analyze_sentiment: Analyze market sentiment for a stock - ticker (string) [required] • generate_recommendation: Generate an investment recommendation report - ticker (string) [required] - analysis (string) [required] # Available Skills (request details via details: {field: 'skills', index: N}): [0] fundamental-analysis - Evaluate a stock's intrinsic value using financial metrics and ratios [1] technical-analysis - Analyze stock price patterns, momentum, and technical indicators [2] portfolio-risk-assessment - Evaluate portfolio risk using volatility, sentiment, and diversification metrics # Context Acquiring If the context contains progressively disclosed information (e.g. skills, history steps) and you want to inspect the details, use the **details** field to request them. The framework will expand these items in the next round. Batch all requests in a single output. When using this field, leave step_content and output empty. ## Field format: - **details**: [{field: "skills", index: 0}, ...] Available fields: **skills** (view a skill's full workflow), **cognitive_history** (view the full result of a previous step) # Output Fields - **step_content**: Your analysis and reasoning for this step - **finish**: Set True when the sub-task is fully complete (default: False) - **details**: Available fields: **skills**, **cognitive_history**. example: [{field: 'skills', index: 0}, ...] - **output**: Tool calls to execute: [{tool, tool_arguments: [{name: 'param', value: 'value'}]}] [18:04:51.260] [Think] (_cognitive_worker.py:332) Message 2 (user, 646 tokens): Based on the context below, decide your next action. Goal: Perform a fundamental analysis of AAPL and give a BUY/HOLD/SELL rating Execution History: [Working Memory (0-2)] [0] Starting fundamental analysis for AAPL. First step is to retrieve financial metrics to evaluate P/E ratio, revenue growth, and debt/equity. Result: results=[ActionStepResult(tool_id='call_0', tool_name='get_financials', tool_arguments={'company': 'AAPL'}, tool_result='AAPL - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45', success=True, error=None)] [1] Retrieving current stock price for AAPL to complete fundamental analysis steps as per procedure. Result: results=[ActionStepResult(tool_id='call_0', tool_name='get_stock_price', tool_arguments={'ticker': 'AAPL'}, tool_result='AAPL: $178.50', success=True, error=None)] [2] Based on the financial metrics: P/E ratio of 28.5 (potentially overvalued), revenue growth of 12% (strong), and debt/equity of 0.45 (healthy). Applying rating logic: P/E >25 but <35, growth >10%, debt <0.7 → HOLD rating. Now generating the recommendation report. Result: results=[ActionStepResult(tool_id='call_0', tool_name='generate_recommendation', tool_arguments={'ticker': 'AAPL', 'analysis': 'AAPL shows strong revenue growth (12%) and healthy debt levels (0.45 D/E), but P/E ratio of 28.5 suggests potential overvaluation. Meets HOLD criteria per fundamental analysis procedure.'}, tool_result='Investment Report for AAPL: AAPL shows strong revenue growth (12%) and healthy debt levels (0.45 D/E), but P/E ratio of 28.5 suggests potential overvaluation. Meets HOLD... Previously Disclosed Details: [skills[0]]: Skill: fundamental-analysis Description: Evaluate a stock's intrinsic value using financial metrics and ratios Instructions: ---------------------------------------- ## Fundamental Analysis Procedure Follow these steps to evaluate a stock: 1. **Get Financial Data**: Retrieve the company's key metrics using get_financials. 2. **Evaluate P/E Ratio**: - P/E < 15: Undervalued (positive signal) - P/E 15-25: Fairly valued - P/E > 25: Potentially overvalued (caution) 3. **Assess Growth**: Revenue growth > 10% is strong, > 20% is exceptional. 4. **Check Debt**: Debt/Equity < 0.5 is healthy, > 1.0 is concerning. 5. **Get Current Price**: Use get_stock_price for the latest quote. 6. **Rating Logic**: - BUY: P/E < 20 AND growth > 10% AND debt/equity < 0.7 - SELL: P/E > 35 OR growth < 0% OR debt/equity > 1.5 - HOLD: Everything else 7. **Generate Report**: Use generate_recommendation with your analysis summary. [18:04:51.260] [Think] (_cognitive_worker.py:332) Total: 1122 tokens (cumulative: 4016) [18:04:51.262] [Think] (_amphibious_automa.py:867) _PromptWorker: finish=True, step=Reviewing the execution history, the fundamental analysis for AAPL has been completed following all required steps: (1) Financial metrics retrieved - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45, (2) Stock price obtained - $178.50, (3) Rating logic applied per procedure - P/E >25 suggests caution, growth >10% is strong, debt <0.5 is healthy, resulting in HOLD rating, (4) Recommendation report generated. All steps of the fundamental-analysis skill procedure have been executed successfully. [18:04:51.262] [Act] (_amphibious_automa.py:873) _PromptWorker: { "content": "Reviewing the execution history, the fundamental analysis for AAPL has been completed following all required steps: (1) Financial metrics retrieved - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45, (2) Stock price obtained - $178.50, (3) Rating logic applied per procedure - P/E >25 suggests caution, growth >10% is strong, debt <0.5 is healthy, resulting in HOLD rating, (4) Recommendation report generated. All steps of the fundamental-analysis skill procedure have been executed successfully.", "result": null, "metadata": { "tool_calls": [] }, "status": null } ================================================== InvestmentAnalyst-e68d4f26 | Completed Tokens: 4016 | Time: 70.92s ================================================== Reviewing the execution history, the fundamental analysis for AAPL has been completed following all required steps: (1) Financial metrics retrieved - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45, (2) Stock price obtained - $178.50, (3) Rating logic applied per procedure - P/E >25 suggests caution, growth >10% is strong, debt <0.5 is healthy, resulting in HOLD rating, (4) Recommendation report generated. All steps of the fundamental-analysis skill procedure have been executed successfully.
What happened?¶
Look at the verbose output above. The LLM saw 3 skill summaries (just names and descriptions). It identified that fundamental-analysis matched its task and requested its details via the details field. The framework then revealed the full procedure, and the LLM followed it step by step — retrieving financials, checking the P/E ratio, evaluating growth, and applying the rating logic.
The other two skills (technical-analysis and portfolio-risk-assessment) were never loaded into context, keeping token usage efficient.
In [8]:
Copied!
# Check which skills were acquired (revealed) during execution
for idx, detail in agent.context.skills._revealed.items():
print(f"[Skill {idx} acquired]")
print(detail[:200] + "...")
# Check which skills were acquired (revealed) during execution for idx, detail in agent.context.skills._revealed.items(): print(f"[Skill {idx} acquired]") print(detail[:200] + "...")
[Skill 0 acquired] Skill: fundamental-analysis Description: Evaluate a stock's intrinsic value using financial metrics and ratios Instructions: ---------------------------------------- ## Fundamental Analysis Procedure...
Part 2: Acquiring History Details¶
The CognitiveHistory also uses LayeredExposure, but for the agent's own execution history. It organizes steps into memory layers:
- Working memory — most recent steps, shown in full detail
- Short-term memory — older steps, shown as summaries only
- Long-term memory — oldest steps, compressed into a paragraph
When an agent has many steps, older ones are pushed into short-term or long-term memory. If the LLM needs specifics from an earlier step, it can use the details field to request them — the same Acquiring mechanism as with skills.
To see this in action, we'll create a CognitiveHistory with a small working memory (working_memory_size=1) so that even a tool calls are enough to push early steps into short-term memory.
In [ ]:
Copied!
from bridgic.amphibious import CognitiveHistory
class ThoroughAnalyst(AmphibiousAutoma[CognitiveContext]):
researcher = think_unit(
CognitiveWorker.inline(
"Gather comprehensive data about the stock. "
"Get price, financials, and sentiment. ",
verbose_prompt=True,
),
max_attempts=6,
)
reporter = think_unit(
CognitiveWorker.inline(
"Generate a final investment recommendation based on all gathered data. "
"If you need details from earlier analysis steps, request them. ",
verbose_prompt=True,
),
max_attempts=3,
)
async def on_agent(self, ctx: CognitiveContext):
await self.researcher
await self.reporter
from bridgic.amphibious import CognitiveHistory class ThoroughAnalyst(AmphibiousAutoma[CognitiveContext]): researcher = think_unit( CognitiveWorker.inline( "Gather comprehensive data about the stock. " "Get price, financials, and sentiment. ", verbose_prompt=True, ), max_attempts=6, ) reporter = think_unit( CognitiveWorker.inline( "Generate a final investment recommendation based on all gathered data. " "If you need details from earlier analysis steps, request them. ", verbose_prompt=True, ), max_attempts=3, ) async def on_agent(self, ctx: CognitiveContext): await self.researcher await self.reporter
In [19]:
Copied!
agent = ThoroughAnalyst(llm=llm, verbose=True)
result = await agent.arun(
goal="Provide a comprehensive investment analysis of AAPL",
tools=[get_stock_price_tool, get_financials_tool, analyze_sentiment_tool, generate_recommendation_tool],
cognitive_history=CognitiveHistory(working_memory_size=1),
)
print(result)
agent = ThoroughAnalyst(llm=llm, verbose=True) result = await agent.arun( goal="Provide a comprehensive investment analysis of AAPL", tools=[get_stock_price_tool, get_financials_tool, analyze_sentiment_tool, generate_recommendation_tool], cognitive_history=CognitiveHistory(working_memory_size=1), ) print(result)
[18:19:49.837] [Router] (_amphibious_automa.py:1573) Auto-detecting execution mode [18:19:49.837] [Router] (_amphibious_automa.py:1579) Detected AGENT mode [18:19:49.837] [Observe] (_amphibious_automa.py:861) _PromptWorker: None [18:19:57.059] [Think] (_cognitive_worker.py:332) Message 1 (system, 363 tokens): Gather comprehensive data about the stock. Get price, financials, and sentiment. Respond in JSON format. # Available Tools (with parameters): • get_stock_price: Fetch stock price data for a ticker symbol - ticker (string) [required] • get_financials: Get financial metrics for a company - company (string) [required] • analyze_sentiment: Analyze market sentiment for a stock - ticker (string) [required] • generate_recommendation: Generate an investment recommendation report - ticker (string) [required] - analysis (string) [required] # Context Acquiring If the context contains progressively disclosed information (e.g. skills, history steps) and you want to inspect the details, use the **details** field to request them. The framework will expand these items in the next round. Batch all requests in a single output. When using this field, leave step_content and output empty. ## Field format: - **details**: [{field: "skills", index: 0}, ...] Available fields: **skills** (view a skill's full workflow), **cognitive_history** (view the full result of a previous step) # Output Fields - **step_content**: Your analysis and reasoning for this step - **finish**: Set True when the sub-task is fully complete (default: False) - **details**: Available fields: **skills**, **cognitive_history**. example: [{field: 'skills', index: 0}, ...] - **output**: Tool calls to execute: [{tool, tool_arguments: [{name: 'param', value: 'value'}]}] [18:19:57.059] [Think] (_cognitive_worker.py:332) Message 2 (user, 35 tokens): Based on the context below, decide your next action. Goal: Provide a comprehensive investment analysis of AAPL Execution History: (none) [18:19:57.059] [Think] (_cognitive_worker.py:332) Total: 398 tokens (cumulative: 398) [18:19:57.062] [Think] (_amphibious_automa.py:867) _PromptWorker: finish=False, step=Starting comprehensive investment analysis for AAPL. I need to gather stock price data, financial metrics, and market sentiment to provide a complete analysis. I'll execute multiple data collection tools in parallel to efficiently gather all required information. [18:19:57.065] [Act] (_amphibious_automa.py:873) _PromptWorker: { "content": "Starting comprehensive investment analysis for AAPL. I need to gather stock price data, financial metrics, and market sentiment to provide a complete analysis. I'll execute multiple data collection tools in parallel to efficiently gather all required information.", "result": { "results": [ { "tool_id": "call_0", "tool_name": "get_stock_price", "tool_arguments": { "ticker": "AAPL" }, "tool_result": "AAPL: $178.50", "success": true, "error": null }, { "tool_id": "call_1", "tool_name": "get_financials", "tool_arguments": { "company": "Apple" }, "tool_result": "Apple - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45", "success": true, "error": null }, { "tool_id": "call_2", "tool_name": "analyze_sentiment", "tool_arguments": { "ticker": "AAPL" }, "tool_result": "Sentiment for AAPL: 65% bullish, 25% neutral, 10% bearish", "success": true, "error": null } ] }, "metadata": {}, "status": null } [18:19:57.066] [Observe] (_amphibious_automa.py:861) _PromptWorker: None [18:20:08.239] [Think] (_cognitive_worker.py:332) Message 1 (system, 363 tokens): Gather comprehensive data about the stock. Get price, financials, and sentiment. Respond in JSON format. # Available Tools (with parameters): • get_stock_price: Fetch stock price data for a ticker symbol - ticker (string) [required] • get_financials: Get financial metrics for a company - company (string) [required] • analyze_sentiment: Analyze market sentiment for a stock - ticker (string) [required] • generate_recommendation: Generate an investment recommendation report - ticker (string) [required] - analysis (string) [required] # Context Acquiring If the context contains progressively disclosed information (e.g. skills, history steps) and you want to inspect the details, use the **details** field to request them. The framework will expand these items in the next round. Batch all requests in a single output. When using this field, leave step_content and output empty. ## Field format: - **details**: [{field: "skills", index: 0}, ...] Available fields: **skills** (view a skill's full workflow), **cognitive_history** (view the full result of a previous step) # Output Fields - **step_content**: Your analysis and reasoning for this step - **finish**: Set True when the sub-task is fully complete (default: False) - **details**: Available fields: **skills**, **cognitive_history**. example: [{field: 'skills', index: 0}, ...] - **output**: Tool calls to execute: [{tool, tool_arguments: [{name: 'param', value: 'value'}]}] [18:20:08.239] [Think] (_cognitive_worker.py:332) Message 2 (user, 236 tokens): Based on the context below, decide your next action. Goal: Provide a comprehensive investment analysis of AAPL Execution History: [Working Memory (0-0)] [0] Starting comprehensive investment analysis for AAPL. I need to gather stock price data, financial metrics, and market sentiment to provide a complete analysis. I'll execute multiple data collection tools in parallel to efficiently gather all required information. Result: results=[ActionStepResult(tool_id='call_0', tool_name='get_stock_price', tool_arguments={'ticker': 'AAPL'}, tool_result='AAPL: $178.50', success=True, error=None), ActionStepResult(tool_id='call_1', tool_name='get_financials', tool_arguments={'company': 'Apple'}, tool_result='Apple - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45', success=True, error=None), ActionStepResult(tool_id='call_2', tool_name='analyze_sentiment', tool_arguments={'ticker': 'AAPL'}, tool_result='Sentiment for AAPL: 65... [18:20:08.239] [Think] (_cognitive_worker.py:332) Total: 599 tokens (cumulative: 997) [18:20:08.242] [Think] (_amphibious_automa.py:867) _PromptWorker: finish=False, step=I have gathered the key data for AAPL: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and sentiment score (65). Now I need to generate a comprehensive investment recommendation based on this collected data. I'll compile the analysis and use the generate_recommendation tool to create the final investment report. [18:20:08.244] [Act] (_amphibious_automa.py:873) _PromptWorker: { "content": "I have gathered the key data for AAPL: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and sentiment score (65). Now I need to generate a comprehensive investment recommendation based on this collected data. I'll compile the analysis and use the generate_recommendation tool to create the final investment report.", "result": { "results": [ { "tool_id": "call_0", "tool_name": "generate_recommendation", "tool_arguments": { "ticker": "AAPL", "analysis": "AAPL Investment Analysis - Stock Price: $178.50 | Financials: P/E Ratio 28.5 (moderate valuation), Revenue Growth 12% (healthy growth), Debt/Equity 0.45 (low leverage, strong balance sheet) | Market Sentiment: 65/100 (moderately positive). Overall assessment shows Apple maintains solid fundamentals with reasonable valuation, consistent revenue growth, and conservative debt levels. Sentiment indicates cautious optimism from market participants." }, "tool_result": "Investment Report for AAPL: AAPL Investment Analysis - Stock Price: $178.50 | Financials: P/E Ratio 28.5 (moderate valuation), Revenue Growth 12% (healthy growth), Debt/Equity 0.45 (low leverage, strong balance sheet) | Market Sentiment: 65/100 (moderately positive). Overall assessment shows Apple maintains solid fundamentals with reasonable valuation, consistent revenue growth, and conservative debt levels. Sentiment indicates cautious optimism from market participants.", "success": true, "error": null } ] }, "metadata": {}, "status": null } [18:20:08.244] [Observe] (_amphibious_automa.py:861) _PromptWorker: None [18:20:15.044] [Think] (_cognitive_worker.py:332) Message 1 (system, 363 tokens): Gather comprehensive data about the stock. Get price, financials, and sentiment. Respond in JSON format. # Available Tools (with parameters): • get_stock_price: Fetch stock price data for a ticker symbol - ticker (string) [required] • get_financials: Get financial metrics for a company - company (string) [required] • analyze_sentiment: Analyze market sentiment for a stock - ticker (string) [required] • generate_recommendation: Generate an investment recommendation report - ticker (string) [required] - analysis (string) [required] # Context Acquiring If the context contains progressively disclosed information (e.g. skills, history steps) and you want to inspect the details, use the **details** field to request them. The framework will expand these items in the next round. Batch all requests in a single output. When using this field, leave step_content and output empty. ## Field format: - **details**: [{field: "skills", index: 0}, ...] Available fields: **skills** (view a skill's full workflow), **cognitive_history** (view the full result of a previous step) # Output Fields - **step_content**: Your analysis and reasoning for this step - **finish**: Set True when the sub-task is fully complete (default: False) - **details**: Available fields: **skills**, **cognitive_history**. example: [{field: 'skills', index: 0}, ...] - **output**: Tool calls to execute: [{tool, tool_arguments: [{name: 'param', value: 'value'}]}] [18:20:15.044] [Think] (_cognitive_worker.py:332) Message 2 (user, 356 tokens): Based on the context below, decide your next action. Goal: Provide a comprehensive investment analysis of AAPL Execution History: (Steps [0-0]: summary only, query details via details) [Short-term Memory (0-0), query details via 'details'] [0] Starting comprehensive investment analysis for AAPL. I need to gather stock price data, financial metrics, and market sentiment to provide a complete analysis. I'll execute multiple data collection tools in parallel to efficiently gather all required information. [Working Memory (1-1)] [1] I have gathered the key data for AAPL: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and sentiment score (65). Now I need to generate a comprehensive investment recommendation based on this collected data. I'll compile the analysis and use the generate_recommendation tool to create the final investment report. Result: results=[ActionStepResult(tool_id='call_0', tool_name='generate_recommendation', tool_arguments={'ticker': 'AAPL', 'analysis': 'AAPL Investment Analysis - Stock Price: $178.50 | Financials: P/E Ratio 28.5 (moderate valuation), Revenue Growth 12% (healthy growth), Debt/Equity 0.45 (low leverage, strong balance sheet) | Market Sentiment: 65/100 (moderately positive). Overall assessment shows Apple maintains solid fundamentals with reasonable valuation, consistent revenue growth, and conservative d... [18:20:15.044] [Think] (_cognitive_worker.py:332) Total: 719 tokens (cumulative: 1716) [18:20:15.048] [Think] (_amphibious_automa.py:867) _PromptWorker: finish=True, step=The comprehensive investment analysis for AAPL is complete. All required data has been gathered: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and market sentiment (65/100). The generate_recommendation tool has been executed with this analysis data to create the final investment report. The task is now complete. [18:20:15.049] [Act] (_amphibious_automa.py:873) _PromptWorker: { "content": "The comprehensive investment analysis for AAPL is complete. All required data has been gathered: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and market sentiment (65/100). The generate_recommendation tool has been executed with this analysis data to create the final investment report. The task is now complete.", "result": null, "metadata": { "tool_calls": [] }, "status": null } [18:20:15.049] [Observe] (_amphibious_automa.py:861) _PromptWorker: None [18:20:22.242] [Think] (_cognitive_worker.py:332) Message 1 (system, 377 tokens): Generate a final investment recommendation based on all gathered data. If you need details from earlier analysis steps, request them. Respond in JSON format. # Available Tools (with parameters): • get_stock_price: Fetch stock price data for a ticker symbol - ticker (string) [required] • get_financials: Get financial metrics for a company - company (string) [required] • analyze_sentiment: Analyze market sentiment for a stock - ticker (string) [required] • generate_recommendation: Generate an investment recommendation report - ticker (string) [required] - analysis (string) [required] # Context Acquiring If the context contains progressively disclosed information (e.g. skills, history steps) and you want to inspect the details, use the **details** field to request them. The framework will expand these items in the next round. Batch all requests in a single output. When using this field, leave step_content and output empty. ## Field format: - **details**: [{field: "skills", index: 0}, ...] Available fields: **skills** (view a skill's full workflow), **cognitive_history** (view the full result of a previous step) # Output Fields - **step_content**: Your analysis and reasoning for this step - **finish**: Set True when the sub-task is fully complete (default: False) - **details**: Available fields: **skills**, **cognitive_history**. example: [{field: 'skills', index: 0}, ...] - **output**: Tool calls to execute: [{tool, tool_arguments: [{name: 'param', value: 'value'}]}] [18:20:22.242] [Think] (_cognitive_worker.py:332) Message 2 (user, 320 tokens): Based on the context below, decide your next action. Goal: Provide a comprehensive investment analysis of AAPL Execution History: (Steps [0-1]: summary only, query details via details) [Short-term Memory (0-1), query details via 'details'] [0] Starting comprehensive investment analysis for AAPL. I need to gather stock price data, financial metrics, and market sentiment to provide a complete analysis. I'll execute multiple data collection tools in parallel to efficiently gather all required information. [1] I have gathered the key data for AAPL: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and sentiment score (65). Now I need to generate a comprehensive investment recommendation based on this collected data. I'll compile the analysis and use the generate_recommendation tool to create the final investment report. [Working Memory (2-2)] [2] The comprehensive investment analysis for AAPL is complete. All required data has been gathered: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and market sentiment (65/100). The generate_recommendation tool has been executed with this analysis data to create the final investment report. The task is now complete. [18:20:22.242] [Think] (_cognitive_worker.py:332) Total: 697 tokens (cumulative: 697) [18:20:37.425] [Think] (_cognitive_worker.py:332) Message 1 (system, 213 tokens): Generate a final investment recommendation based on all gathered data. If you need details from earlier analysis steps, request them. Respond in JSON format. # Available Tools (with parameters): • get_stock_price: Fetch stock price data for a ticker symbol - ticker (string) [required] • get_financials: Get financial metrics for a company - company (string) [required] • analyze_sentiment: Analyze market sentiment for a stock - ticker (string) [required] • generate_recommendation: Generate an investment recommendation report - ticker (string) [required] - analysis (string) [required] # Output Fields - **step_content**: Your analysis and reasoning for this step - **finish**: Set True when the sub-task is fully complete (default: False) - **output**: Tool calls to execute: [{tool, tool_arguments: [{name: 'param', value: 'value'}]}] [18:20:37.425] [Think] (_cognitive_worker.py:332) Message 2 (user, 1026 tokens): Based on the context below, decide your next action. Goal: Provide a comprehensive investment analysis of AAPL Execution History: (Steps [0-1]: summary only, query details via details) [Short-term Memory (0-1), query details via 'details'] [0] Starting comprehensive investment analysis for AAPL. I need to gather stock price data, financial metrics, and market sentiment to provide a complete analysis. I'll execute multiple data collection tools in parallel to efficiently gather all required information. [1] I have gathered the key data for AAPL: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and sentiment score (65). Now I need to generate a comprehensive investment recommendation based on this collected data. I'll compile the analysis and use the generate_recommendation tool to create the final investment report. [Working Memory (2-2)] [2] The comprehensive investment analysis for AAPL is complete. All required data has been gathered: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and market sentiment (65/100). The generate_recommendation tool has been executed with this analysis data to create the final investment report. The task is now complete. Previously Disclosed Details: [cognitive_history[0]]: Content: Starting comprehensive investment analysis for AAPL. I need to gather stock price data, financial metrics, and market sentiment to provide a complete analysis. I'll execute multiple data collection tools in parallel to efficiently gather all required information. Result: results=[ActionStepResult(tool_id='call_0', tool_name='get_stock_price', tool_arguments={'ticker': 'AAPL'}, tool_result='AAPL: $178.50', success=True, error=None), ActionStepResult(tool_id='call_1', tool_name='get_financials', tool_arguments={'company': 'Apple'}, tool_result='Apple - P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45', success=True, error=None), ActionStepResult(tool_id='call_2', tool_name='analyze_sentiment', tool_arguments={'ticker': 'AAPL'}, tool_result='Sentiment for AAPL: 65% bullish, 25% neutral, 10% bearish', success=True, error=None)] [cognitive_history[1]]: Content: I have gathered the key data for AAPL: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and sentiment score (65). Now I need to generate a comprehensive investment recommendation based on this collected data. I'll compile the analysis and use the generate_recommendation tool to create the final investment report. Result: results=[ActionStepResult(tool_id='call_0', tool_name='generate_recommendation', tool_arguments={'ticker': 'AAPL', 'analysis': 'AAPL Investment Analysis - Stock Price: $178.50 | Financials: P/E Ratio 28.5 (moderate valuation), Revenue Growth 12% (healthy growth), Debt/Equity 0.45 (low leverage, strong balance sheet) | Market Sentiment: 65/100 (moderately positive). Overall assessment shows Apple maintains solid fundamentals with reasonable valuation, consistent revenue growth, and conservative debt levels. Sentiment indicates cautious optimism from market participants.'}, tool_result='Investment Report for AAPL: AAPL Investment Analysis - Stock Price: $178.50 | Financials: P/E Ratio 28.5 (moderate valuation), Revenue Growth 12% (healthy growth), Debt/Equity 0.45 (low leverage, strong balance sheet) | Market Sentiment: 65/100 (moderately positive). Overall assessment shows Apple maintains solid fundamentals with reasonable valuation, consistent revenue growth, and conservative debt levels. Sentiment indicates cautious optimism from market participants.', success=True, error=None)] [cognitive_history[2]]: Content: The comprehensive investment analysis for AAPL is complete. All required data has been gathered: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and market sentiment (65/100). The generate_recommendation tool has been executed with this analysis data to create the final investment report. The task is now complete. Metadata: tool_calls: [] [18:20:37.425] [Think] (_cognitive_worker.py:332) Total: 1239 tokens (cumulative: 1936) [18:20:37.428] [Think] (_amphibious_automa.py:867) _PromptWorker: finish=True, step=All required data for AAPL investment analysis has been collected and processed. The generate_recommendation tool was successfully executed in step 1, producing a comprehensive investment report. The analysis includes stock price ($178.50), financial metrics (P/E 28.5, 12% revenue growth, 0.45 debt/equity), and market sentiment (65/100). The final recommendation is ready based on this synthesized data. [18:20:37.428] [Act] (_amphibious_automa.py:873) _PromptWorker: { "content": "All required data for AAPL investment analysis has been collected and processed. The generate_recommendation tool was successfully executed in step 1, producing a comprehensive investment report. The analysis includes stock price ($178.50), financial metrics (P/E 28.5, 12% revenue growth, 0.45 debt/equity), and market sentiment (65/100). The final recommendation is ready based on this synthesized data.", "result": null, "metadata": { "tool_calls": [] }, "status": null } ================================================== ThoroughAnalyst-034580c6 | Completed Tokens: 3652 | Time: 47.59s ================================================== All required data for AAPL investment analysis has been collected and processed. The generate_recommendation tool was successfully executed in step 1, producing a comprehensive investment report. The analysis includes stock price ($178.50), financial metrics (P/E 28.5, 12% revenue growth, 0.45 debt/equity), and market sentiment (65/100). The final recommendation is ready based on this synthesized data.
In [20]:
Copied!
# Inspect the history layers
history = agent.context.cognitive_history
print(f"Total steps: {len(history)}")
print(f"Working memory size: {history.working_memory_size}")
print()
for line in history.summary():
print(line)
# Inspect the history layers history = agent.context.cognitive_history print(f"Total steps: {len(history)}") print(f"Working memory size: {history.working_memory_size}") print() for line in history.summary(): print(line)
Total steps: 4 Working memory size: 1 [Short-term Memory (0-2), query details via 'details'] [0] Starting comprehensive investment analysis for AAPL. I need to gather stock price data, financial metrics, and market sentiment to provide a complete analysis. I'll execute multiple data collection tools in parallel to efficiently gather all required information. [1] I have gathered the key data for AAPL: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and sentiment score (65). Now I need to generate a comprehensive investment recommendation based on this collected data. I'll compile the analysis and use the generate_recommendation tool to create the final investment report. [2] The comprehensive investment analysis for AAPL is complete. All required data has been gathered: stock price ($178.50), financial metrics (P/E: 28.5, Revenue Growth: 12%, Debt/Equity: 0.45), and market sentiment (65/100). The generate_recommendation tool has been executed with this analysis data to create the final investment report. The task is now complete. [Working Memory (3-3)] [3] All required data for AAPL investment analysis has been collected and processed. The generate_recommendation tool was successfully executed in step 1, producing a comprehensive investment report. The analysis includes stock price ($178.50), financial metrics (P/E 28.5, 12% revenue growth, 0.45 debt/equity), and market sentiment (65/100). The final recommendation is ready based on this synthesized data.
What happened?¶
We configured CognitiveHistory(working_memory_size=1), meaning only the 1 most recent steps are shown in full detail. The rest are pushed into short-term memory as summaries.
The agent executed in two phases: researcher gathered data (price, financials, sentiment), then reporter generated the final recommendation. Because working memory is so small, the researcher's early tool calls are already in short-term memory by the time the reporter starts.
If the reporter needed specifics from an earlier research step (e.g., the exact P/E ratio from get_financials), it could use the details field to request the full content — exactly like acquiring skill details.
Rehearsal — Mental Simulation Before Acting¶
Rehearsal lets the agent mentally simulate its planned action before committing. This is especially valuable for high-risk operations — the agent can catch potential issues before they happen.
With rehearsal enabled, before committing to a tool call, the LLM can fill a rehearsal field to simulate what would happen. The framework processes this simulation round and then asks the LLM to make its final decision.
Use cases: irreversible actions, financial decisions, operations with significant side effects.
# Enable rehearsal on a think unit
analyst = think_unit(
CognitiveWorker.inline(
"Analyze the stock carefully before recommending.",
enable_rehearsal=True,
),
max_attempts=8,
)
Reflection — Assessing Information Quality¶
Reflection lets the agent evaluate whether the information it has gathered is sufficient and consistent before acting. The LLM fills a reflection field to assess data quality, and the framework injects the assessment into the next round.
Use cases: unreliable data sources, contradictory signals, multi-source cross-referencing.
# Enable reflection on a think unit
analyst = think_unit(
CognitiveWorker.inline(
"Assess data consistency before recommending.",
enable_reflection=True,
),
max_attempts=8,
)
Combining Policies¶
All three policies can be enabled simultaneously. They fire in order: Acquiring → Rehearsal → Reflection. Each fires at most once per OTC cycle. This gives your agent the maximum deliberation depth.
analyst = think_unit(
CognitiveWorker.inline(
"Perform a comprehensive analysis.",
enable_rehearsal=True,
enable_reflection=True,
# Acquiring is always enabled by default
),
max_attempts=10,
)
Policy Rules:
- Each policy fires at most once per observe-think-act cycle, then closes
- Policies fire in order: Acquiring → Rehearsal → Reflection
- After all active policies have fired, the LLM must commit to a final action
- Acquiring is built-in and always available; Rehearsal and Reflection must be explicitly enabled
- Acquiring is disabled when
output_schemais set on the worker
What have we learnt?¶
In this tutorial, we explored how cognitive policies add multi-round deliberation to agents within a single OTC cycle:
- Acquiring (built-in, always active): Enables progressive disclosure through
LayeredExposurefields. The LLM sees summaries first and can request full details on demand.- Skill details: The agent sees skill names/descriptions, requests the full procedure only when needed — keeping many skills available without flooding context.
- History details: Older execution steps are summarized into memory layers. The agent can request specifics from any earlier step.
- Rehearsal (
enable_rehearsal=True): Lets the LLM mentally simulate its planned action before committing. Ideal for high-risk operations. - Reflection (
enable_reflection=True): Lets the LLM assess information quality and consistency before acting. Useful when data may be unreliable. - All policies fire at most once per OTC cycle, in the order Acquiring → Rehearsal → Reflection.
- Policies can be combined — enable multiple policies on the same
CognitiveWorkerfor maximum deliberation depth.