chan-trend-mechanism

name	chan-trend-mechanism
description	Use when analyzing A-share or index OHLCV data with multi-timeframe Chan-style structure, support/pressure ladders, MACD auxiliary context, current-state observations, or deterministic condition branches without action language.

Chan Trend Mechanism

Overview

Use this skill to turn local, closed-bar OHLCV parquet data into a deterministic multi-timeframe structure reading. The skill describes current structure only: background, operating level, support/pressure ladder, auxiliary MACD context, and if/then condition branches.

Hard Boundaries

• Do not use private commentary sources, scraping interfaces, session files, credentials, screenshots, or exported posts.
• Do not use action language such as buy, sell, add, reduce, signal, alpha, or edge in generated conclusions.
• Do not let MACD or any indicator override the structural posture. MACD is only auxiliary momentum context.
• Do not infer unavailable bars. Emit missing-data warnings and keep the conclusion narrower.

Workflow

1. Inspect the user's available data path and symbol universe.
2. If MACD columns are missing and the user wants momentum context, run scripts/macd_enrichment.py.
3. For one symbol, run scripts/structure_analyzer.py.
4. For multiple indices or symbols, run scripts/panel.py.
5. Use the JSON as canonical; treat Markdown as deterministic presentation.

Example commands, replacing the skill directory with the actual local path:

python skills/chan-trend-mechanism/scripts/macd_enrichment.py \
  --data-dir data/index/derived \
  --timeframes 1d 60m 120m

python skills/chan-trend-mechanism/scripts/structure_analyzer.py \
  --symbol 000001.SH \
  --name 上证指数 \
  --as-of "2026-05-15 15:00:00" \
  --data-dir data/index/derived \
  --output-dir output/chan_states

python skills/chan-trend-mechanism/scripts/panel.py \
  --symbols 000001.SH 399001.SZ 399006.SZ 000688.SH \
  --as-of "2026-05-15 15:00:00" \
  --data-dir data/index/derived \
  --output-dir output/chan_panel

Output Contract

The analyzer emits chanlun_structure_v0.2 JSON with these core sections:

• background: daily and weekly style context from 1d, 2d, 1w, 2w.
• operating_level: current operating structure from 30m, 60m, 120m.
• support_ladder: MA20/MA55 support and pressure across all configured timeframes.
• momentum_context.macd: fixed 12/26/9 MACD facts for 1d, 60m, 120m.
• conditional_tree: mechanical condition branches.
• current_state_description: short current-state paragraphs for human reading.

Interpretation Rules

• Use 1d/2d/1w/2w to describe background.
• Use 30m/60m/120m to locate the operating level.
• Use MA20 as the midline and MA55 as the next structural line.
• Use MACD phrases only as mechanical facts: zero-axis location, DIF/DEA cross, histogram expansion/shrinkage/flip.
• Prefer "current observation", "condition", "support", "pressure", "structure improves", and "structure weakens" over any action-oriented wording.

References

• Read references/data_schema.md when adapting external data into the expected parquet layout.
• Read references/mechanism.md when writing or reviewing conclusions from the JSON output.