Market Data

Normalize and summarize market data for: $ARGUMENTS

Overview

Implementation status: code-backed
Local entry script: <bundle-root>/market-data/run.py
Primary purpose: expose the canonical one-symbol market-data pipeline without adding decision, thesis, or execution claims
Workflow stages: stage 2 Data Collection & Quality Assurance plus stage 3 Data Cleaning & Normalization, with stage-4 feature-base preparation
Local executor guarantee: schema normalization, required-column validation, date sorting, duplicate-date collapse, numeric coercion, indicator enrichment, and optional export

Use When

The user wants to inspect whether a symbol’s raw price data is usable for research or backtesting.
The user wants an enriched export that downstream skills or external workflows can reuse.
The user wants to verify date coverage, source, row count, and the basic feature base before analysis.
The caller needs a disciplined preprocessing step before market-analyze, stock-data, market-screen, or backtest-evaluator.

Do Not Use When

The user wants scored market interpretation. Use market-analyze.
The user wants a reusable multi-block research snapshot. Use stock-data.
The user wants action guidance or execution planning. Use decision-support or strategy-design.
The user wants provider refresh orchestration or stale-record diagnosis across the broader runtime. Use data-sync.

Inputs

Normal case: one stock symbol.
Optional --csv PATH: load from a local CSV or JSON file rather than the default data provider.
Optional --export PATH: write the enriched dataset to a target path.
Optional --start, --end, --source: constrain the data-loading path.
If symbol is omitted, the skill may reuse last_symbol from the same execution context.
Local file contract:
- minimum normalized columns required after alias resolution: date, open, high, low, close, volume
- optional supported columns include amount, pct_change, turnover, and name
- the agent should state whether the file is externally supplied, bundle-generated, or host-generated
Adjustment note:
- the runtime uses the configured adjustment mode when pulling provider data
- the agent should say whether the dataset is provider-fetched or external-file-backed when the adjustment basis matters

Execution

Step 1: Define the acquisition scope

Before running the local executor, identify:

the symbol or explicit local file being normalized
the intended date window
the expected downstream use: inspection, export, research snapshot, screening, or backtest preparation
whether point-in-time correctness matters for the next stage

Step 2: Run the local executor

python3 <bundle-root>/market-data/run.py <symbol> [--csv PATH] [--export PATH]

Step 3: Review the normalized dataset as a research input, not just a table

The agent should check and report:

source provenance: akshare or local file
date coverage and row count
duplicate-date handling and obvious missing-window issues
whether required columns were present directly or inferred through alias normalization
whether turnover or amount are absent, because that weakens later liquidity commentary
whether the requested date window is too short for downstream indicators or backtests

Step 4: Confirm the feature base and known limits

The local enrichment path adds moving averages, MACD, RSI, Bollinger bands, ATR, volume ratio, support, resistance, drawdown, and breakout proximity. The agent should state:

that these are convenience features for downstream skills, not validated alpha factors by themselves
that the pipeline is daily-bar oriented rather than intraday
that local normalization does not by itself solve point-in-time corporate-action, suspension, or survivorship questions for backtesting

Output Contract

Minimum local executor output: human-readable text beginning with 数据处理.
Core fields: symbol, source, date range, row count, last close, last volume, turnover, optional export path.
Local executor side effects:
- updates session history for the current execution context
- writes an enriched dataset only when --export is present
Caller-facing delivery standard:
- explicitly label the dataset source and whether the symbol or file was reused from session context
- say whether the dataset is being used for inspection, export, screening, or backtest preparation
- if a local file was supplied, treat its provenance and adjustment basis as user-supplied assumptions
- identify any material quality issues that affect downstream use, especially short history, missing turnover, suspicious gaps, unsupported columns, stale files, or external-file ambiguity
Non-guarantees:
- no persisted run directory is created by this skill
- no claim is made about strategy validity, signal strength, or execution readiness

Failure Handling

Parse and argument errors: non-zero exit with a readable 命令错误 message.
Unsupported or malformed market data: readable failure text beginning with 执行失败:.
Missing required columns in a local file: readable failure text rather than a traceback.
Missing symbol with no reusable session symbol: readable guidance instead of a traceback.
If the dataset is technically readable but analytically weak, the skill should still return with explicit quality warnings rather than pretending the input is backtest-ready.

Key Rules

Use this skill for data inspection and preprocessing, not recommendation.
Distinguish local executor guarantees from broader research hygiene that the agent must still review.
Do not treat enriched indicators as a substitute for point-in-time feature engineering discipline.
For backtests or retrospective evaluation, explicitly state whether the dataset appears clean enough for the intended holding horizon and realism assumptions.

Composition

This skill is the data foundation for market-analyze, stock-data, market-screen, decision-dashboard, decision-support, strategy-design, and backtest-evaluator.
data-sync is the upstream orchestration surface when freshness or provider health is the real task.