0.1.8 • Published yesterdayCLI

a-share-after-hours-brief-skill

Licence

MIT

Version

0.1.8

Deps

Size

205 kB

Vulns

Weekly

Summary Dependency Versions

中文说明

This Codex / Agent Skills package reviews one or more A-share stocks after the market closes.

It does not pick stocks, trade, or write a full-market recap. It takes the stock or stock pool named by the user, creates a phone-friendly HTML brief, and saves local JSON history so the next run can check whether the previous observation points held up.

When to use it

Review one A-share stock after the close and see whether today's move changed the thesis.
Review a small stock pool by comparing price action, events, and follow-up conditions.
Check the last review and mark each observation point as verified, partly verified, not verified, invalidated, or unable to judge.
Create an HTML brief for archiving, sharing, or email attachment.
Review position discipline when the user provides holdings or transaction records.

Core features

Reviews one stock or a small stock pool.
Verifies the previous review with verified / partially_verified / not_verified / invalidated / unable_to_judge.
Saves structured history as JSON in a history/ folder next to the HTML output.
Uses market, sector, and supply-chain context to separate market, industry, upstream/downstream, and stock-specific factors.
Handles announcements, earnings, meetings, policy changes, industry news, and upstream/downstream signals.
Separates assessability from direction. When evidence is sufficient, it assigns 向上 / 维持震荡 / 向下; when evidence is insufficient, it records that explicitly instead of defaulting to sideways.
Calculates local K-line features from daily OHLCV data, including returns, moving averages, trend structure, candle location, volume state, range/breakout structure, volatility, gaps, and relative context.
Keeps absolute technical structure separate from benchmark and sector relative performance, so relative outperformance does not overwrite the stock's own price-volume structure.
Computes Pearson correlation from multi-stock K-line returns by default and beta against a named benchmark when benchmark K-line data is available.
Uses a modern flat HTML template made for phone reading and email attachments. Direction tags follow the A-share color convention: red for up, green for down.
Can create a short Gmail draft body when the user asks for one. HTML attachment support is checked separately before claiming an attachment was added.
Checks position discipline only when the user provides holdings, trade records, or asks for it directly.

Scope limits

It does not provide default buy or sell recommendations.
It does not execute trades.
It does not provide deterministic price forecasts or target prices.
Local K-line features are evidence for the review, not a standalone forecast or trading signal.
It does not generate a full market-wide recap by default.
It does not create weekly reviews or long-term memory by default.

Installation

npm / npx

npx a-share-after-hours-brief-skill install

Default install location:

~/.codex/skills/a-share-after-hours-brief/

Overwrite an existing installation:

npx a-share-after-hours-brief-skill install --force

--force first moves the existing installation to a timestamped backup directory. Use --force --no-backup only when you explicitly want to overwrite without a backup.

Install to a custom skills directory:

npx a-share-after-hours-brief-skill install --target /path/to/skills

GitHub

If npm is not available, install directly from GitHub:

npx github:SkyBridgeM/a-share-after-hours-brief-skill install

Manual

From the repository root, copy the skill files into the expected skill directory:

mkdir -p ~/.codex/skills/a-share-after-hours-brief
cp -R SKILL.md agents assets config references scripts schemas examples \
  ~/.codex/skills/a-share-after-hours-brief/

Usage examples

Review today's CATL after market close

Generate an after-hours HTML brief for these two A-share stocks

Verify the previous review and list next-trading-day observation conditions

Review this A-share watchlist and create a Gmail draft

Chinese prompts work as well:

复盘今天的宁德时代

校验上次判断，列出下一交易日验证条件

Output

The generated HTML review usually includes:

Today's conclusion
Concise market background
Previous-review verification
Stock-pool overview
Individual stock review cards
Major events
Industry and supply-chain news
Correlation analysis
Next-session assessment and validation conditions
Optional position discipline review

The next-session section first decides whether the available evidence is assessable. If it is, the report chooses one of 向上 / 维持震荡 / 向下 and shows the confidence level. If evidence is insufficient, the report says so and does not force a sideways judgment. It is not a trading instruction and does not include a target price.

K-line output separates two layers: absolute technical structure and relative-performance context versus benchmark or sector. Data-quality limitations are shown by feature group, especially when volume data or comparison data is incomplete.

JSON History

By default, history is saved next to the HTML output:

reports/
├── 2026-06-16__a-share-after-hours__300750-SZ_600519-SH.html
└── history/
    └── 2026-06-16__300750-SZ_600519-SH.json

History rules:

JSON is the only structured history source.
The skill does not generate a Markdown history log.
The JSON record does not store absolute paths.
generated_at uses ISO 8601 with a timezone offset. A-share reports default to Asia/Shanghai.
The report directory can be moved or backed up as a folder.
Running the same stock pool again on the same day updates the matching JSON record.
Malformed history files are skipped with structured warnings instead of being silently ignored.
The formal schema lives at schemas/history-v1.schema.json; a sample record is available at examples/history-record.example.json.

File naming

Use stable ASCII filenames for new local artifacts and keep Chinese titles inside the HTML or Markdown content. The directory names are configurable; the pattern standardizes filenames, not a single mandatory storage path.

Artifact	Pattern
HTML report	`<report-output-dir>/YYYY-MM-DD__a-share-after-hours__<sorted-stock-codes>.html`
History JSON	`<history-dir>/YYYY-MM-DD__<sorted-stock-codes>.json`
Run state / logs	`<state-or-log-dir>/YYYY-MM-DD.<purpose>.json`, `.txt`, or `.log`
Raw run data	`<data-dir>/YYYY-MM-DD-data/<code>_<kind>.json` or another dated `data` / `raw_data` path
Monthly summary	`monthly-summary-YYYY-MM.html`, `.md`, or `.json`

Example filename: 2026-06-25__a-share-after-hours__002745-SZ_301307-SZ.html.

Local storage cleanup

The skill includes a local cleanup helper for report folders that grow over time. It previews the cleanup by default and deletes files only when --apply is passed.

Default retention:

Type	Retention
HTML reports	30 days
Charts	30 days
Raw data	14 days
Cache	7 days
Logs	14 days
History JSON	90 days
Monthly summaries	Keep forever

Preview cleanup:

python3 scripts/cleanup_reports.py --root ./reports

For older workspaces that keep artifacts in several places, such as report HTML in the workspace root, history/ at the root, raw data folders, or run-state/log directories, run the preview against the workspace root instead:

python3 scripts/cleanup_reports.py --root .

Apply cleanup:

python3 scripts/cleanup_reports.py \
  --root ./reports \
  --policy config/storage-policy.example.json \
  --apply

The script prints the files it would delete, the estimated space that would be freed, and the files it skipped with reasons. It never deletes current-month files. History JSON older than 90 days is deleted only when a monthly summary for the same month already exists, so daily review records are not removed before a monthly archive is in place. Dated JSON/TXT/LOG files in obvious run-state or log directories, such as logs/, state/, run-state/, runtime/, or automation-state/, are treated as logs. Dated data, raw_data, or *-data artifacts are treated as raw data. Ordinary business JSON outside those artifact directories is not deleted by default.

Repository structure

.
├── SKILL.md
├── agents/
│   └── openai.yaml
├── assets/
│   ├── brief-template.html
│   └── plain-email-summary-template.md
├── config/
│   └── storage-policy.example.json
├── examples/
│   └── history-record.example.json
├── references/
│   ├── data-providers.md
│   ├── event-triggers.md
│   ├── history-and-review.md
│   ├── html-email.md
│   ├── industry-news.md
│   ├── kline-analysis.md
│   ├── review-discipline.md
│   ├── storage-cleanup.md
│   └── wind-data.md
├── schemas/
│   └── history-v1.schema.json
└── scripts/
    ├── cleanup_reports.py
    ├── correlation.py
    ├── kline_features.py
    └── review_journal.py

Scripts

Calculate local K-line structure features:

python3 scripts/kline_features.py stock.json

Calculate K-line features with benchmark and sector relative strength:

python3 scripts/kline_features.py stock.json \
  --benchmark benchmark.json \
  --sector sector.json \
  --adjustment forward

Calculate return correlation for multiple stocks, with optional beta against a benchmark:

python3 scripts/correlation.py stock_a.json stock_b.json stock_c.json \
  --labels StockA StockB StockC \
  --benchmark benchmark.json \
  --benchmark-label CSI300

Look up the previous history record:

python3 scripts/review_journal.py lookup \
  --history-dir ./reports/history \
  --before-date 2026-06-16 \
  --stocks 300750.SZ,600519.SH

Build the current history record and save it:

python3 scripts/review_journal.py build \
  --input current-draft.json \
  --output-html ./reports/2026-06-16_A-share-after-hours-brief.html

Preview local storage cleanup:

python3 scripts/cleanup_reports.py --root ./reports

Delete the listed files after reviewing the dry-run output:

python3 scripts/cleanup_reports.py --root ./reports --apply

Requirements

Python 3.10+
Node.js 18+, only for the npm installer
A Codex / Agent Skills compatible client
Financial data capabilities for A-share quotes, adjusted K-lines, announcements, financial facts, benchmark context, and news. Wind MCP is preferred when available.
Agent Reach is optional. It can supplement external web, industry association, upstream/downstream, and peer news searches.
Gmail connector, only when Gmail draft creation is requested

The bundled Python scripts use only the standard library.

Local Checks

npm run check

You can also check the Python scripts directly:

PYTHONPYCACHEPREFIX=/tmp/a-share-after-hours-brief-skill-pycache \
python3 -m py_compile scripts/review_journal.py scripts/correlation.py scripts/kline_features.py scripts/cleanup_reports.py

The full check also runs Python unit tests and Node.js installer tests using only standard-library and built-in runtime features.

If you have the Skill Creator validator:

python3 /path/to/skill-creator/scripts/quick_validate.py ./a-share-after-hours-brief

Data and privacy

Generated HTML reports, history/*.json, raw data, cache files, holding records, and trade records stay in the user's local output folder. The skill does not upload them or store them in a central service. Use scripts/cleanup_reports.py when you want to review or prune old local artifacts.

This repository only includes reusable skill logic, templates, and documentation. It does not include Wind, Gmail, or other API credentials.

Publishing

Package name:

a-share-after-hours-brief-skill

Install command:

npx a-share-after-hours-brief-skill install

GitHub repository:

https://github.com/SkyBridgeM/a-share-after-hours-brief-skill