research-goal · diff
git:20261003.8a2bc5b to git:20261028.ec06a0f
2 added, 1 removed. Audit A to A.
What this update changes in the checks
No check changed: every fixed check gives the same answer before and after this update.
A tripped check is a label, not a verdict: read the matched text and the lines below. This week's updates that newly trip a check.
---
name: research-goal
description: Goal-driven finance research workflow: attach a research-only objective, track criteria, and add evidence while avoiding live trading execution.
category: flow
---
# 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. When all required criteria have been audited, call `update_research_goal_status`.
+ 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
- Keep evidence short and concrete.
- Prefer artifact-backed evidence when a tool produced a run or file.
- Include `run_id`, `artifact_path`, `source_provider`, `source_type`, `symbol_universe`, `benchmark`, and `data_as_of` when known.
- For evidence backed by one specific tool result, copy that result's exact `tool_call_id` into `add_goal_evidence`; do not shorten, alias, or invent it.
- `evidence_id` values such as `ev_...` identify rows in the research-goal ledger. They are used by goal audits and are never grounding source refs; do not write `ev_...::field` in a figures declaration.
- `tool_call_id` preserves provenance but does not verify goal completion by itself. Completion still needs verified evidence from an existing `run_id` or an allowed `artifact_path` with a matching sha256 hash.
- Do not mark live trading instructions as evidence. Refuse or reframe them as research-only analysis.
## Completion Rules
- Use `update_research_goal_status(status="complete")` only after every required criterion has an audit row.
- Satisfied audit rows must cite verified `evidence_ids`.
- Use `status="blocked"` or `status="insufficient_evidence"` when evidence is missing, stale, contradictory, or not verifiable.
- Use `status="cancelled"` only when the user explicitly asks to end or discard the goal.
## Response Style
Tell the user what the goal is tracking only when it helps. Do not flood the answer with ledger details. Lead with the research conclusion, then mention which criteria remain unresolved.