committee

/committee - Investment Committee (Data-Driven Edition)

Have Claude, Codex, and Gemini independently analyze the same set of objective data, cross-validate when necessary, and ultimately extract actionable consensus.

How to Use

/committee - Convene the investment committee
/committee 红利ETF该怎么操作 - Convene for a specific question
/committee consensus - Summarize the existing views of the three models

Execution Steps (Enhanced Version Recommended)

0. Select Mode

Simple: one round of independent analysis → summarize consensus
Enhanced: R1 independent analysis → R2 cross-validation → summarize consensus
Default: ask the user; if unspecified, use the enhanced version

Step 1: Prepare Unified Input Data

Obtain market data

cd "股市信息" && python3 scripts/fetch_market_data.py

The script already supports technicals(MA/RSI/MACD/range position). If it is too slow, you can turn it off in MODULES .

Read context Read 股市信息/Config/Context.md, and supplement with Profile.md、Principles.md、Insight.md 's risk preferences and constraints
Determine today's decision question Ask the user what question they want the committee to discuss today
Generate input file Integrate the above content and save it to 股市信息/Committee/Input/YYYY-MM-DD-Input.md

Use template:Committee/Templates/prompt_template.md
Must include a "Data Quality Checklist", clearly specifying missing fields (valuation/capital flow/technical, etc.)
fetch_market_data.py In the holdings/watchlist output, technicals and trading data must be included

Step 2: Round1 Independent Analysis

Based on the input data, output according to Committee/Templates/opinion_template.md :

Overall market judgment
Position signal matrix and action recommendations (each target recommendation must cite >=2 objective data fields)
Risk warnings
Answer the decision question
Confidence self-assessment

Save to:

Committee/Opinions/Claude.md
Committee/Opinions/Codex.md
Committee/Opinions/Gemini.md

Step 3: Cross-Validation (Enhanced Version)

Provide the R1 views to the other models, requiring them to:

Point out omissions/misreadings/over-inferences based only on the same data
Allow adjustments to: recommendations, confidence, risk warnings
Output an R2 revision table

Save to:

Committee/Opinions/Claude-R2.md
Committee/Opinions/Codex-R2.md
Committee/Opinions/Gemini-R2.md

Step 4: Summarize Consensus

When the user indicates that all views have been collected:

Read the view files Prefer R2; if unavailable, use R1
Compare target by target

3/3 in agreement → strong consensus
2/3 in agreement → weak consensus
All different → disagreement

Weighting and risk control rules

Weight = confidence × data completeness
Views that do not cite objective data are downweighted
Any model's high-risk warning must be included in the summary

Generate consensus report According to Templates/consensus_template.md output and save to Committee/Sessions/YYYY-MM-DD.md

Step 5: Record Recommendations to Be Tracked

Record strong consensus recommendations in 股市信息/Config/Insight.md 's "Adoption Record"
Set the status to "Pending Adoption", and later use /trade to verify

Output Format Requirements

Uniformly use Committee/Templates/opinion_template.md
Missing fields must be marked as "missing", and filling them in on your own is prohibited
Key recommendations must include a "key data basis"

Consensus extraction rules (supplementary)

If there is inconsistent understanding of key data (e.g., different readings of RSI/trend), enter the "disagreement zone" and indicate the data dispute
If the data gap is relatively large, downgrade the original strong consensus to a weak consensus
If a "risk red line" appears (e.g., RSI > 75 and the price is far above MA60), caution must be advised and confidence lowered

Template file location

Input prompt template:Committee/Templates/prompt_template.md
Opinion output template:Committee/Templates/opinion_template.md
Consensus summary template:Committee/Templates/consensus_template.md