AGENT SKILL

thesis-tracker

Create and maintain investment/trading theses as living pages with explicit key_bets that resolve over time. The structural-conviction layer above individual trades. Feeds the market_call calibration domain.

License 许可
MIT
所属仓库
singhvedant/tbrain
最近更新
2026-07-04
Stars
★ 0
Forks
0

01 前言 Front Matter

name
thesis-tracker
triggers
[ "new thesis", "I think the market", "my view is", "I'm bullish on", "I'm bearish on", "track this thesis", "update my thesis on" ]
description
Create and maintain investment/trading theses as living pages with explicit key_bets that resolve over time. The structural-conviction layer above individual trades. Feeds the market_call calibration domain.
writes_pages
[ "theses/*", "macro/*" ]

03 SKILL.md 正文

thesis-tracker — living theses with resolvable bets

A thesis is a directional or structural view that outlives any single trade ("AI capex cycle runs through 2027", "INR bonds roll over"). Write it as a thesis page (or macro page for regime-level views) carrying explicit key_bets[] — falsifiable sub-claims that resolve later via post-mortems. This is what feeds the market_call calibration domain: it measures whether the operator's high-conviction calls are actually right.

When to invoke

  • "New thesis: …" / "my view is …" / "I'm bullish/bearish on …"
  • The operator states a multi-week+ directional belief
  • Updating an existing thesis as evidence arrives (append, don't overwrite)

Frontmatter

---
type: thesis
thesis_text: "AI datacenter capex compounds through 2027; memory + power are the bottlenecks."
instruments: [nvda, mu, vrt]   # thesis_for edges
sectors: [semiconductors, power]
market_view: bullish           # bullish | bearish | neutral | volatility
vintage: 2026-06
invalidation: "Two consecutive hyperscaler capex guide-downs."
key_bets:
  - id: bet-1
    claim: "NVDA data-center revenue grows >40% YoY in FY27."
    confidence: 0.7
  - id: bet-2
    claim: "Memory pricing stays firm through H1-2027."
    confidence: 0.55
---

Body

State the reasoning, the evidence for and against, and the second-order exposures. Link instruments/sectors by name so the graph wires thesis_for and in_sector edges.

How to write / update

gbrain capture --file <thesis.md>          # new
gbrain query "thesis on <topic>"           # find existing before creating

Before creating, ALWAYS gbrain query for an existing thesis on the topic — update it rather than forking a duplicate. Append dated evidence under a ## Updates section.

After writing

  • Each key_bet becomes a bet-kind take. When reality resolves a bet, run trade-postmortem (or a dedicated resolution) to record resolved_outcome
    • learned_pattern against the bet's take_id.
  • Surface contradictions: run gbrain think "what contradicts <thesis>" and note opposing evidence in the thesis body so the dream cycle's contradiction detector has signal.
  • A trade citing this thesis (via thesis: frontmatter) links back automatically — no manual cross-linking needed.

04 同仓库更多 Skills