Stock Sentiment Analysis

Stock Sentiment Analysis

Use this skill as the shared sentiment layer for market skills. It does not fetch data by itself; it tells Codex how to interpret evidence gathered by cn-stock-move-reason, jp-stock-move-reason, stock-technical-analysis, us-stock-gamma-moomoo, public news, forums, breadth, and user-provided screenshots or notes.

This public-safe skill must not contain personal information, API keys, account data, private paths, raw screenshots, full copied notes, ticker-specific personal trade logs, or proprietary labels from a private RAG corpus. It must remain usable without any private RAG folder.

Public And Private Versions

If both public and private versions of a market skill exist, prefer the private version for local analysis when the user permits it. Use this public skill as the public-safe shared framework and as the release checklist source.

When updating a paired private/public skill, write public-safe generalized lessons to both versions, but keep private paths, private labels, raw notes, screenshots, account data, and personal trade context only in the private version or private RAG index.

When preparing a GitHub upload or public release, use only the public version and read the repo-level release/privacy check from the repository root at shared/references/release-and-privacy.md first. Run a privacy check for private RAG folders, .ftindex, .env, credentials, personal paths, raw source files, screenshots, and private labels.

Reference Reading Rule

When this skill selects a reference file, first scan the file structure, then read the sections and nearby guardrails relevant to the current task. Read the complete file when it is short, when the task is broad or strategic, or when partial reading could miss constraints. Do not rely on stale memory or heading-only scans. For publishing, still read the repo-level shared/references/release-and-privacy.md carefully enough to complete the checklist.

Workflow

1. Read the relevant areas of references/experience.md before deep analysis; read the full file when the task is broad or the active playbook may affect the answer.
2. During decomposition, actively expand stock, index, and theme questions into emotion-cycle, main-line/follower, expectation-gap, crowding, cross-market sentiment, or mainline/funds/game/cycle checks when these lenses can change the conclusion, even if the user did not explicitly request them. For collective sector surges, continuation questions, and market-sector-stock resonance, apply Market-Sector-Stock Resonance And Continuation from references/sentiment-framework.md; separate logic durability, tape continuity, and entry quality instead of treating a strong narrative as an automatic buy point. Read the relevant areas of references/sentiment-framework.md. Read the repo-level shared/references/release-and-privacy.md before publishing, syncing public/private versions, or preparing a GitHub upload.
3. Gather or receive evidence from the market-specific skill first:
   • A-shares: prefer cn-stock-move-reason for quote, announcements, 股吧, board ranks, breadth, and A-share emotion cycle.
   • Japanese stocks: prefer jp-stock-move-reason for quote, news, Yahoo 掲示板, metrics, and theme/peer context.
   • US stocks/indexes: prefer us-stock-move-reason for why-up/why-down/mover questions, choose us-stock-gamma-moomoo for option/gamma questions, and choose stock-technical-analysis for chart/trend questions; use them together when catalyst, positioning, and price action all matter.
   • For U.S. community discussion evidence, moomoo-comment-sentiment can be used when installed. Treat it as a moomoo community sample that helps measure retail heat, disagreement, chasing, and panic; it is not a full-market sentiment survey and must not replace news, filings, earnings, option positioning, or price behavior.
   • Pre-screened narrative-status entries from market reports can be used as social-media-derived clues for current themes, mainline candidates, and crowding/expectation-gap questions. Account/source quality can be assumed acceptable for screening, but timeliness is mandatory: stale source/update times are background only, not current sentiment evidence.
4. For A-share evidence, optional 东方财富妙想 skills can supplement the market-specific workflow when installed. MX data is an evidence and screening layer, not a replacement for the existing know-how: still apply source hierarchy, emotion-cycle staging, main-line/follower judgment, expectation-gap analysis, forum/news psychology, breadth, sector rotation, macro, and technical confirmation when relevant. Use mx-data for quote/financial/fund-flow/sector data, mx-search for news/announcements/research/policy, and mx-xuangu for sector constituents, concept stocks, peer screens, and natural-language condition screens. For A-share questions such as 这个板块有哪些股票, 相关股, 概念股, 龙头股, 板块成分, or 同题材还有谁, try mx-xuangu first when available; then use mx-data/mx-search selectively to classify purity, heat, and catalysts. If 妙想 is unavailable, continue with public sources or state the limitation. You may briefly suggest installing/configuring 妙想 only when it would materially improve the exact request; never make it a dependency.
5. Do not use account-touching 妙想 skills automatically. Use mx-zixuan only when the user explicitly asks to query/add/delete/filter 东方财富 self-selected stocks; for 自选股里哪些符合条件, first try mx-xuangu constrained to self-selected stocks, and if unsupported, combine mx-zixuan self-selected results with mx-xuangu screening locally. Use mx-moni only for explicit simulated-portfolio queries or simulated trades.
6. Classify the move through three lenses: confirmed catalyst, emotion/positioning, and technical confirmation. When resonance matters, finish the hierarchy market -> sector/theme -> stock and test broad-market support, turnover/liquidity, participation breadth, industry-chain diffusion, continuing fundamental validation, and lifecycle position. Do not let forum heat replace confirmed news.
7. If a multi-turn correction reveals a reusable lesson, update references/experience.md after answering. Generalize the lesson; remove ticker-specific personal details and private labels.

DTM Context

Prefer data already obtained by the upstream market-specific skill. When the relevant context is missing, use the canonical JSON interfaces in https://daytrading.monster/api-docs/: https://daytrading.monster/api/themes for cross-market theme members and completed-session participation/relative strength; https://daytrading.monster/api/chinastock-anomaly for A-share theme rotation, limit-up structure, and move reasons; and https://daytrading.monster/api/24hfeed/x-monitor for discussion within the returned eight-hour snapshot window. Read themes[] and constituents[], including coverage and quote dates; do not infer live flows from daily returns or whole-market sentiment from an account sample. Parse the text/plain bodies as JSON. Fetch only the layers needed for the question, retaining the existing source hierarchy and interpretation rules.

Optional Private RAG

If the user wants to use private study materials, ask them to specify a local RAG or index folder. Do not assume a default private path. Use it only for extracting reusable rules relevant to the current task.

Rules for private RAG:

• Never copy raw private files, full passages, screenshots, API keys, account data, or personal trade logs into this public skill.
• Never write private folder paths into public files.
• Strip proprietary strategy names, private person names/handles, and memorable private labels.
• Summarize only public-safe, reusable principles into references/experience.md or references/sentiment-framework.md when the user explicitly asks to update the skill.
• Keep private indexes outside the Git repository. A user may maintain their own local index such as private-rag/ or any folder they choose; it must be ignored by Git.
• When useful, offer to build or update a local index for the user's private RAG. The index should contain only file names or user-approved aliases, topics, page/slide ranges, keywords, and short public-safe summaries. It should not copy raw source text or include secrets.

Local RAG Index Pattern

If the user asks to use private study materials repeatedly, propose a lightweight index stored outside the public repository. The same private RAG can support sentiment, technical analysis, gamma/option analysis, and cross-market study materials:

private-rag-index/
  market-sentiment-framework.md
  technical-patterns.md
  gamma-notes.md
  macro-and-cross-market.md

Each index entry should be compact:

- Topic: high-level divergence
  Source: user-private-file alias + page/slide range
  Keywords: climax, failed breakout, volume without progress
  Reusable rule: Late-cycle divergence after one-sided euphoria is riskier than early-cycle turnover.
  Public-safe: yes/no

Use the index to find relevant private material efficiently, then answer from public-safe distilled rules.

Output Style

When used directly, answer in Chinese unless the user asks otherwise. Use this compact structure when useful:

1. 情绪结论: risk-on/risk-off, early-cycle/late-cycle, panic, rotation, or crowded long.
2. 证据: confirmed news, forum/post heat, breadth, sector peers, option/gamma positioning, and chart behavior.
3. 周期位置: A-share seven-stage cycle when relevant; otherwise describe low-vol accumulation, early breakout, acceleration, distribution, or de-risking.
4. 主线判断: leader/follower/defensive alternative/old-leader rebound/noise.
5. 验证条件: what confirms continuation.
6. 失效条件: what shows emotion has turned.

Do not give direct trading instructions. Give conditional conclusions and clearly label uncertain forum narratives as 思惑 or 未确认.