Market Brief

Generate the default one-symbol A-share brief for: $ARGUMENTS

Overview

• Implementation status: code-backed
• Local entry script: <bundle-root>/market-brief/run.py
• Primary purpose: deliver the default first-pass one-symbol synthesis by combining current market state, conditional action framing, and conditional execution planning into one persisted run
• Research layer: composed skill spanning descriptive analysis, decision support, and execution planning
• Workflow stages: stage 4 State Interpretation, stage 6 Decision Support, and stage 7 Execution Planning in one fast bundled pass
• Design intent: speed and completeness over depth; single-pass synthesis without thesis pressure-testing
• Local executor guarantee: generate the persisted composed brief and artifact set from the current snapshot plus the shared decision and strategy helpers

Use When

• The user wants the default one-symbol entry point.
• The user wants one bounded output covering current state, what the system would do, and how it would structure the trade.
• The user wants persisted artifacts that downstream skills can inspect or evaluate later.
• The user prioritizes speed over depth and does not need thesis pressure-testing, variant views, or invalidation logic.
• The user wants a complete picture quickly without asking follow-up questions.

Do Not Use When

• The user only wants raw normalized data. Use market-data.
• The user only wants descriptive state reading without action or execution layers. Use market-analyze.
• The user asks "why," "what drives this," "what are the risks," or "what would change your view." Use analysis.
• The user needs a research memo suitable for investment committee review or retrospective evaluation. Use analysis.
• The user wants thesis framing, catalyst analysis, variant views, disconfirming evidence, or invalidation conditions. Use analysis.
• The user wants stored artifact retrieval or comparison rather than fresh computation. Use reports or analysis-history.
• The user wants batch ranking or watchlist triage. Use market-screen or decision-dashboard.

Routing Boundary: market-brief vs. analysis

This is the most important routing decision in the bundle. Use this operational rule:

Use market-brief when:

• User wants current positioning quickly
• User accepts single-pass synthesis without pressure-testing
• User does not ask about drivers, risks, or what could go wrong
• Speed matters more than depth

Use analysis when:

• User asks "why," "what drives this," "what are the risks," "what would change your view"
• User needs a research memo for IC review or formal evaluation
• User wants saved artifacts for later retrospective evaluation with backtest-evaluator
• User explicitly requests depth, thesis framing, or risk pressure-testing
• The decision has material portfolio impact and requires documented thesis discipline

When in doubt: If the user's question implies they need to understand causation, risks, or invalidation conditions, route to analysis. If they just want to know current state and what to do about it, use market-brief.

Inputs

• Normal case: one stock symbol.
• Optional --csv PATH: use a local CSV instead of the default market source.
• Optional --start, --end, --source: constrain the data-loading path.
• Optional --capital, --cash, --position, --risk, --max-position: shape the decision-support block.
• Optional --style, --hold-days: shape the execution-planning block.
• Optional --notify: attempt outbound Feishu delivery after report creation.
• If symbol is omitted, the skill may reuse last_symbol from the same execution context.
• Important assumption boundary: if cash, position, risk, style, or hold-days are omitted, local defaults drive the decision and strategy sections. The caller should surface those defaults as assumptions rather than treating them as user-approved constraints.

Execution

Step 1: Confirm this is the right layer mix

Use market-brief only when the user actually wants descriptive analysis, decision support, and execution planning in one pass. If the user wants only one layer or wants deeper research pressure-testing, route accordingly.

Critical routing check: Does the user's question imply they need thesis-level rigor? If yes, route to analysis instead.

Step 2: Run the local executor

python3 <bundle-root>/market-brief/run.py <symbol> [--csv PATH] [--capital N] [--position N] [--notify]

Step 3: Deliver the result with labeled layers

Keep 数据处理, 市场分析, 决策支持, and 策略设计 distinct in the caller-facing answer. Make clear that the last two are conditional heuristics, not executable orders.

Layer separation requirements:

• Data layer: state the data source, date window, and whether symbol was reused from session
• Analysis layer: present as descriptive current-state interpretation, not as thesis validation
• Decision layer: frame as conditional guidance with explicit assumptions (see decision-support contract for three-part structure)
• Execution layer: frame as conditional planning anchors, not as guaranteed fill levels

Step 4: Acknowledge what market-brief does not provide

When delivering the output, explicitly state:

• "This is a single-pass synthesis without thesis pressure-testing"
• "For deeper analysis including variant views, disconfirming evidence, and invalidation conditions, use the analysis skill"
• If the user later asks "why" or "what are the risks," recommend escalating to analysis

Step 5: Reuse the run artifacts downstream

If the next step is export, comparison, or retrospective evaluation, pass the generated run directory to reports, analysis-history, or backtest-evaluator instead of recomputing blindly.

Output Contract

• Minimum local executor output: human-readable text beginning with # <symbol> A 股简报 plus four labeled blocks: 数据处理, 市场分析, 决策支持, and 策略设计.
• Artifact side effects: writes one dated run directory with state.json, report.md, and metadata.json.
• Caller-facing delivery standard:
  • Layer visibility: keep descriptive analysis separate from action sizing and execution planning
  • Data provenance: identify the market-data source, date window, and whether the symbol was supplied explicitly or reused from session context
  • Assumption transparency: state any defaulted capital, cash, position, risk, style, or hold-period assumptions that materially shaped the answer
  • Confidence framing: treat score and confidence as heuristic decision aids, not forecast probabilities
  • Risk disclosure: surface non-modeled risks when relevant, especially catalyst gaps, liquidity constraints, opening gap risk, price-limit behavior, and stop slippage
  • Conditional framing: present action and plan blocks as advisory frames, not as executable orders
  • Scope acknowledgment: explicitly state that this is a single-pass synthesis without thesis pressure-testing
  • Escalation path: if the user asks follow-up questions about drivers, risks, or invalidation, recommend analysis
• Optional side effect: if --notify is used, the skill attempts Feishu notification without blocking the report.

Failure Handling

• Parse and argument errors: non-zero exit with a readable 命令错误 message.
• Runtime data errors: readable failure text beginning with 执行失败:.
• Missing symbol with no reusable session symbol: readable guidance instead of a traceback.
• Notification delivery problems: fail open; the report still returns.
• If optional context is absent, do not imply the brief is fully informed on fundamentals, catalysts, or news.

Key Rules

• Prefer this skill as the default one-symbol entry point, but do not let default status turn it into a catch-all.
• Treat this as a first-pass synthesis, not as an IC-grade deep dive by itself.
• If the user asks what would change the view, recommend escalating to analysis for proper invalidation logic.
• Treat all action and plan blocks as advisory framing rather than executable trading instructions.
• The value of market-brief is speed and completeness, not depth. When depth is needed, route to analysis.
• Layer separation is mandatory. Do not let the composed output blur into a single undifferentiated recommendation.

Composition

• Builds on the same shared loaders and scoring logic used by market-data, market-analyze, decision-support, and strategy-design.
• Produces artifacts that should be reused by reports, analysis-history, and backtest-evaluator.
• analysis is the escalation path when the caller needs stronger thesis discipline than the default brief provides.
• Should not be used when the user's question implies they need thesis-level rigor, variant views, or invalidation logic.