Goal-Driven Finance Research

Use this skill when a user asks for a multi-step finance research task, comparison, audit, thesis review, or "keep working until the answer is supported." The goal runtime is for research only. Never use it to place, submit, or execute trades.

When to Attach a Goal

Start a goal when the task has any of these traits:

• It needs multiple criteria before a conclusion is credible.
• It compares strategies, assets, regimes, or evidence sources.
• It may continue across turns or require a final audit.
• The user asks for "long driven task", "goal", "审计", "对比", "研究结论", or similar.

Do not start a goal for a tiny one-shot answer unless the user explicitly asks.

Tool Flow

1. Call start_research_goal with a concise research-only objective.
2. Include 3-5 acceptance criteria when the user provided enough context.
3. Before continuing an existing task, call get_research_goal.
4. After a market-data lookup, backtest, document read, web source, or manual reasoning step, call add_goal_evidence.
5. Link evidence to a criterion using criterion_id or criterion_index.
6. If the objective wording needs tightening or correcting mid-research, call update_research_goal instead of starting a new goal; it edits the objective in place and keeps existing criteria and evidence.
7. When all required criteria have been audited, call update_research_goal_status.

Criteria Template

Use this shape when the user did not provide criteria:

• Define the research-only thesis and symbol universe.
• Collect fresh market, benchmark, or artifact evidence.
• Compare alternatives or a control baseline.
• Record caveats, contradictions, and the non-advice boundary.

Evidence Rules