# Stock Portfolio Brief — Decide What Changed and What to Do

Turn recorded US stock holdings and fresh sources into a plain-language decision: hold, buy, trim, switch or wait. It compares the proposed action with doing nothing and an appropriate index, and makes missing evidence visible.

## When it fires

Use for portfolio reviews, daily stock briefs, earnings thesis checks and stock tips. This is not a general news digest, a trading bot or an always-on service.

## Install

```sh
npx skills add m1nga/daily-brief
```

In Codex: `Use $daily-brief to review my stock holdings and tell me what changed.`

Provide a portfolio CSV with `ticker,shares,avg_cost` and your investment constraints, or just name a stock to start research. Current cash is optional; without it the skill does not invent buying power or whole-account weights. The host needs browsing for fresh filings. The Python 3 collector uses only the standard library and a public Yahoo endpoint; availability is not guaranteed. It has no API-key requirement and no brokerage or messaging client.

## What it does

A lead decision, dated evidence, strongest counterargument, a feasible conditional plan when supported, and the next event to review. Numerical snapshots and briefs are saved separately so old beliefs can be compared with new evidence.

Fictional example: two shares at $80 cost, a verified $100 quote and $50 cash produce $250 net assets, $40 unrealized gain and an 80% holding weight. Without a cash balance, net assets and weight stay unknown. A missing quote never becomes a 100% loss or a clean bill of health.

## Design notes

The earlier workflow repeatedly reused an old stock thesis in new daily reports and conflated unavailable checks with passing checks. This version separates arithmetic from source-based judgment and treats unknown as a real state. It does not force model scores, investor-persona debates or daily trade quotas.

The machine name `daily-brief` is retained for compatibility. Public instructions contain no private account, broker, credential or portfolio defaults. The skill does not create schedules or send reports without a separate request.

## Validation

The bundled collector is covered by nine executed Python unittest cases for account math, missing cash and quotes, stale or invalid evidence, flat accounts and malformed CSVs. Live collection was exercised separately; that verifies endpoint access at test time, not future availability or investment performance. Instruction routing and financial-period failure cases were manually simulated, not evaluated by independent users. Tests live in `tests/test_snapshot.py`.

## Author

Created and maintained by [Ming](https://github.com/m1nga). Released under the repository MIT license. Research support only; the user makes and executes investment decisions.