Analysis

Analysis

Run a deeper one-symbol analysis workflow for: $ARGUMENTS

Overview

• Implementation status: code-backed
• Local entry script: /analysis/run.py
• Primary purpose: build a research memo that enforces institutional thesis discipline through mandatory structural elements: explicit thesis statement, identified key drivers, articulated variant views, disconfirming evidence, and invalidation conditions
• Research layer: deeper memo that may compose descriptive, decision, and execution layers, but must pressure-test them rather than simply restate them
• Workflow stages: stage 1 Strategy Framing, stage 4 Signal and State Interpretation, and stage 6 Decision Context at memo depth
• Local executor guarantee: generate a persisted research packet and artifact set that the agent can elevate into a thesis-disciplined memo

Use When

• The user asks "why," "what drives this," "what are the risks," or "what would change your view."
• The user wants a research memo suitable for investment committee review or retrospective evaluation.
• The user needs saved artifacts that will later feed analysis-history, reports, or backtest-evaluator.
• The user explicitly requests depth, thesis framing, catalyst analysis, or risk pressure-testing.
• The user wants to understand not just current state but what could go wrong or what would invalidate the current view.

Do Not Use When

• The user wants the default concise full-stock brief. Use market-brief.
• The user only wants direct action, quantity, or stop and take-profit anchors. Use decision-support.
• The user only wants execution zones and checklist output. Use strategy-design.
• The user wants artifact retrieval or comparison rather than fresh computation. Use reports or analysis-history.
• The user wants batch triage rather than one-symbol depth. Use market-screen or decision-dashboard.
• The available evidence is too thin to support thesis-level claims. In this case, either narrow to market-brief or acknowledge evidence gaps explicitly.

Inputs

• Normal case: one stock symbol.
• Optional market inputs: --csv, --start, --end, --source.
• Optional account inputs: --capital, --cash, --position, --risk, --max-position.
• Optional planning inputs: --style, --hold-days, --depth compact|standard|deep, --strategy STRATEGY_ID (repeatable).
• Optional news inputs: --headline, --headline-file, --company-name, --alias, --max-items.
• If symbol is omitted, the skill may reuse last_symbol from the same execution context.
• Important assumption boundary: any benchmark, scenario, or portfolio assumption that comes from the user prompt rather than the local executor must be labeled as a user-supplied assumption in the final memo.

Execution

Step 1: Confirm evidence sufficiency for thesis-level analysis

Before proceeding, assess whether the available evidence supports thesis-level claims. Thesis-level analysis requires:

• Sufficient price history to identify trend structure and regime
• Observable technical or fundamental drivers that can be articulated specifically
• Enough context to frame at least one plausible alternative interpretation

If evidence is insufficient, either:

• Route to market-brief for a descriptive synthesis without thesis claims, or
• Proceed with analysis but explicitly narrow the thesis section to acknowledge evidence gaps

Step 2: Run the local executor

python3 /analysis/run.py  [--headline-file PATH] [--depth standard] [--strategy STRATEGY_ID]

Step 3: Structure the memo with mandatory thesis discipline

The analysis memo must include these structural elements when evidence permits:

Thesis Statement (mandatory):

• One clear sentence stating the investment view
• Must be falsifiable: a thesis that cannot be wrong is not a thesis
• Example: "600519 is positioned for a trend continuation based on institutional accumulation and sector rotation tailwinds"

Key Drivers (mandatory, 3-5 maximum):

• Specific, observable factors supporting the thesis
• Ranked by importance
• Each driver must be concrete enough to monitor
• Example: "1) Net institutional inflow over 10 consecutive sessions, 2) Sector relative strength vs. benchmark, 3) Technical breakout above 200-day resistance"

Primary Catalyst (mandatory when identifiable):

• The single most important near-term event or condition that would validate or accelerate the thesis
• Must have observable timing or trigger conditions
• If no clear catalyst exists, state that explicitly: "No identifiable near-term catalyst; thesis depends on continuation of current technical structure"

Variant View (mandatory):

• At least one alternative interpretation of the same evidence
• Must be intellectually honest, not a strawman
• Should articulate why a reasonable investor might reach a different conclusion
• Example: "Alternative view: current strength reflects late-cycle momentum exhaustion rather than institutional accumulation; volume profile suggests retail participation rather than smart money"

Disconfirming Evidence (mandatory):

• Specific observations that weaken or contradict the thesis
• Must be current and material, not hypothetical
• If no material disconfirming evidence exists, state that explicitly but acknowledge this is unusual
• Example: "Disconfirming: sector breadth is narrowing, with only 30% of sector constituents above their 50-day MA despite index strength"

Invalidation Conditions (mandatory):

• Specific, observable conditions that would falsify the thesis
• Must include both price-based and non-price-based conditions
• Should be monitorable in real-time or near-real-time
• Example: "Thesis invalidated if: 1) price closes below 180 support on volume, 2) sector relative strength turns negative for 3+ sessions, 3) institutional flow reverses to net selling"

Evidence Gaps (mandatory when material):

• Explicit acknowledgment of what is not known or not observable
• Should state what additional information would strengthen or weaken the thesis
• Example: "Evidence gaps: no visibility into institutional mandate changes, limited fundamental disclosure on Q4 margins, unclear regulatory timeline for sector policy"

Step 4: Separate layers explicitly

When the memo includes decision support or execution planning layers, keep them visibly distinct:

• Use clear section headers or transitions
• Restate the account or style assumptions that drive those layers
• Do not let action or execution framing masquerade as thesis validation

Step 5: Preserve the artifact trail

Reference the generated run directory when handing off to reports, analysis-history, or backtest-evaluator. The artifact should be complete enough to support retrospective evaluation without requiring conversational context.

Output Contract

• Minimum local executor output: human-readable text beginning with # Analysis.
• Minimum local sections: Data Perspective, Intelligence, Battle Plan, and Workflow Blueprint when available.
• Artifact side effects: writes one dated run directory with state.json, report.md, and metadata.json.
• Caller-facing delivery standard:
• Mandatory thesis structure: every analysis memo must include thesis statement, key drivers, primary catalyst (or explicit statement that none exists), variant view, disconfirming evidence, and invalidation conditions
• Evidence sufficiency: if any mandatory element cannot be supported by available evidence, state that gap explicitly rather than fabricating content
• Provenance separation: distinguish computed facts from heuristic interpretation, optional enrichments from core analysis, and user-supplied assumptions from local defaults
• Manual context handling: if manual headlines are used, label them as user-supplied or file-backed context and explicitly state they are not independently verified
• Confidence framing: treat numeric confidence fields as heuristic conviction scores, never as statistical hit-rate estimates or calibrated probabilities
• Layer visibility: keep the distinction between descriptive analysis, decision support, and execution planning visible; do not collapse the memo into a disguised order ticket
• Non-modeled risks: when decision or execution layers are included, surface material non-modeled risks including catalyst uncertainty, liquidity constraints, gap risk, limit behavior, and stop slippage

Failure Handling

• Parse and argument errors: non-zero exit with a readable 命令错误 message.
• Market-data failures: readable failure text beginning with 执行失败:.
• Optional fundamentals degrade fail-open through the shared helper.
• If --headline-file is unreadable or malformed, the command returns readable 执行失败: text.
• Evidence-thin scenarios: if the available evidence cannot support thesis-level claims, either:
• Route to market-brief for descriptive synthesis without thesis structure, or
• Proceed with analysis but explicitly acknowledge evidence gaps in each mandatory section
• Never pad evidence-thin sections with generic narrative or fabricated conviction

Key Rules

• Depth means better pressure-testing, not just more words. The value of analysis over market-brief is structural thesis discipline, not length.
• Thesis structure is mandatory, not optional. If evidence cannot support a thesis element, acknowledge the gap explicitly.
• Variant views must be intellectually honest. Do not create strawman alternatives just to fill the section.
• Invalidation conditions must be specific and monitorable. Vague conditions like "if the market changes" are not acceptable.
• Preserve explicit provenance for any news, fundamental, or user-supplied context.
• Do not present heuristic confidence or battle-plan outputs as order instructions.
• Route to decision-support or strategy-design when the user wants the narrower action or execution layer rather than the full memo.

Composition

• Builds on market-data, market-analyze, decision-support, strategy-design, and optional news-intel logic.
• Produces artifacts that can feed reports, analysis-history, and backtest-evaluator.
• Often serves as the deepest one-symbol artifact before retrospective evaluation.
• Should be the default choice when the user's question implies they need thesis-level rigor rather than just current-state description.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: NatsuFox
• Source: NatsuFox/A-Stockit
• License: MIT

Install and usage instructions live in the source repository linked above.