"Vibe-Trading: Your Personal Trading Agent"
Python ai-agentalgorithmic-tradingbacktestingfintechllmmcpmulti-agentpythonquantitative-financetrading
fix(swarm): attribute committee verdicts to the framework, not the person
Haozhe Wu8b56e6ed94db
diff --git a/agent/src/swarm/presets/value_investing_committee.yaml b/agent/src/swarm/presets/value_investing_committee.yamlindex 5962c94ee..200d675bf 100644--- a/agent/src/swarm/presets/value_investing_committee.yaml+++ b/agent/src/swarm/presets/value_investing_committee.yaml@@ -1,6 +1,6 @@name: value_investing_committeetitle: "Value Investing Committee (Buffett / Munger / Duan / Li Lu)"-description: "Four adversarial master perspectives — Buffett (moat & price), Munger (inversion & risk), Duan Yongping (good business & trustworthy management), Li Lu (10-year certainty & civilizational trend, incl. demographic/consumption macro) — run in parallel on one company, then a chair synthesizes consensus and contradictions into a verdict. Inspired by the ai-berkshire master-视角 framework. Use for deep value-investing deliberation on a single named company."+description: "Four adversarial master perspectives — Buffett (moat & price), Munger (inversion & risk), Duan Yongping (good business & trustworthy management), Li Lu (10-year certainty & civilizational trend, incl. demographic/consumption macro) — run in parallel on one company, then a chair synthesizes consensus and contradictions into a verdict. Inspired by the ai-berkshire master-视角 framework. Each perspective is a reconstruction of a publicly documented methodology, not the named individual and not an endorsement. Use for deep value-investing deliberation on a single named company."agents:- id: buffett@@ -137,13 +137,14 @@ agents:## You must output1. **Consensus** (all four agree; highest confidence) — e.g. "the moat is real", "management is trustworthy", "valuation is stretched".- 2. **Contradictions** (who disagrees with whom, on what) — e.g. Buffett thinks cheap, Munger sees a path to zero; Duan likes management, Li Lu doubts 10-year certainty. **Explain the root cause of the disagreement** (caliber difference / time-horizon difference / risk-preference difference).+ 2. **Contradictions** (which lens disagrees with which, on what) — e.g. the moat-and-price lens reads the price as cheap while the inversion lens finds a path to zero; the good-business lens trusts management while the 10-year-certainty lens doubts durability. **Explain the root cause of the disagreement** (caliber difference / time-horizon difference / risk-preference difference).3. **Verdict** — on consensus + contradictions, give invest / watch / avoid + confidence (information sufficiency).4. **Fatal risks & "thesis-broken" signals** — synthesize Munger's failure paths into 3 stop-loss triggers.5. **One-page decision table** — core logic 3 sentences / value range / key assumptions & validation nodes / conclusion.## Principles- - Adversarial tension is by design: Buffett says "cheap enough", Munger asks "how could it die", Duan says "good business", Li Lu asks "still here in 10 years" — when synthesizing, don't smooth contradictions; explain them.+ - Adversarial tension is by design: the moat-and-price lens asks "cheap enough", the inversion lens asks "how could it die", the good-business lens asks "would you entrust management with your money", the 10-year-certainty lens asks "still here in 10 years" — when synthesizing, don't smooth contradictions; explain them.+ - **Attribute to the framework, never to the person.** Write "under the moat-and-price framework, the price is defensible", never "Buffett would buy this". The four upstream analyses are reconstructions of publicly documented methodologies — not the opinions of the named individuals, not written or endorsed by them. State this once in the report, and state that the verdict is analysis, not investment advice.- Honestly report information gaps: if the four masters lack information on some dimension, state "under missing X, conclusion confidence is Y".- Use report_audit (extract → verdict) as a quality gate on key numbers in the final report.