Guide · Agent skill
How to use the research-goal agent skill for finance research
Updated · written with AI from the project's own documentation and checked against our catalogue data
research-goal is an agent skill from the Vibe-Trading project that makes an AI agent treat a multi-step finance research question as a tracked goal. It is written for analysts, researchers and developers who want each conclusion linked to the evidence behind it. After this guide you will know what the skill changes, how to install it, how to trigger it, and what it needs to run.
Goal-Driven Finance ResearchHKUDS/Vibe-Trading · MIT Download the skill Quick start
- 1Get the Vibe-Trading runtime. Confirm you have the Vibe-Trading agent runtime that provides the four research-goal tools. The skill cannot run without them.
- 2Download and unzip the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or .claude/skills/ for one project.
- 3Or upload it in the Claude apps. Open Settings → Capabilities → Skills and upload the skill there if you use the Claude apps.
- 4Ask a multi-step research question. Describe a comparison, audit or thesis review in plain language. Mention a goal or audit if you want the tracked workflow.
- 5Provide criteria if you have them. List 3-5 acceptance criteria. If you do not, the agent can use the built-in template.
- 6Read the conclusion and open criteria. Check the research conclusion first, then the list of criteria that remain blocked or lack evidence.
What an agent skill is and what this one changes
An agent skill is a folder containing a SKILL.md file that tells an AI agent when and how to handle a certain kind of task. The agent reads the file when the task matches the skill's description.
With research-goal installed, a complex finance question is handled as a goal. The agent writes a research-only objective, sets acceptance criteria (ideally 3-5), and links each step such as a market data lookup, backtest, document read or web source to the criterion it supports. It marks the goal complete only when every required criterion has supporting evidence. Otherwise it reports the goal as blocked or lacking evidence.
- Tracks a research-only objective and its criteria
- Links evidence to specific criteria by criterion_id or criterion_index
- Reports unresolved criteria instead of guessing
- Never places, submits or executes trades
Install the research-goal skill
Download the ZIP of the skill from this page and unzip it. For a personal install, place the unzipped folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.
In the Claude apps you can instead upload the skill under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload
Check the requirements before you rely on it
The skill does not work on its own. It depends on the Vibe-Trading agent runtime, which provides the start_research_goal, get_research_goal, add_goal_evidence and update_research_goal_status tools. Those tools are not described in the SKILL.md file itself, so the skill's instructions have nothing to call without that runtime.
The skill file includes no scripts and makes no network calls.
- Requires: the Vibe-Trading agent runtime with the four goal tools
- Runs scripts: no
- Uses network: no
Trigger the skill with natural requests
You do not need a special command. Ask for a multi-step task in plain language. The skill is meant for requests such as a comparison, an audit, a thesis review, or a task you want kept going until the answer is supported. Using words like "goal", "audit" or "comparison" helps the agent recognise the task.
Do not expect a goal for a quick one-shot question. The skill says not to open one unless you ask for it explicitly.
- "Compare two momentum strategies on the same universe and tell me which criteria are still unresolved."
- "Audit my thesis that small-cap value outperformed in the last regime. Keep working until the answer is supported."
- "Review this research note and list which claims lack evidence."
What the agent does during a goal
The agent calls start_research_goal with a concise objective, then gets the goal with get_research_goal before continuing an existing task. After each lookup, backtest, document read, web source or manual reasoning step, it adds evidence and links it to a criterion. When every required criterion has been audited, it updates the status.
If you give no criteria, the skill proposes a template: define the thesis and symbol universe, collect fresh market or benchmark evidence, compare against a control baseline, and record caveats and the non-advice boundary.
- Start: start_research_goal with a research-only objective
- Resume: get_research_goal
- Record: add_goal_evidence, linked to a criterion
- Close: update_research_goal_status with complete, blocked, insufficient_evidence or cancelled
Read the results and know the limits
The agent leads with the research conclusion and then names the criteria that remain open. Completion requires verified evidence from an existing run_id or an allowed artifact_path with a matching sha256 hash. A tool_call_id alone does not prove completion.
The skill sets a non-advice boundary, so outputs are research analysis, not investment recommendations. It refuses instructions to place or execute live trades and does not treat them as evidence.
- Status complete only when every required criterion has audited evidence
- Use blocked or insufficient_evidence when evidence is missing, stale or contradictory
- Output is research analysis, not advice
What people use it for
- Comparing two or more trading strategies on the same symbol universe with a control baseline
- Checking whether an investment thesis holds across market regimes, with each claim tied to evidence
- Reviewing a research note and listing which claims still lack supporting data
- Building a repeatable finance research workflow on the Vibe-Trading platform
Frequently asked questions
Is research-goal free to use?
The source file does not state a price or licence terms beyond the directory listing, which records the licence as MIT. Check the repository licence for the full terms. The skill also depends on the Vibe-Trading runtime, which this file does not describe.
What do I need to install research-goal?
You need the Vibe-Trading agent runtime that provides start_research_goal, get_research_goal, add_goal_evidence and update_research_goal_status. Then download the skill ZIP and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
Does research-goal place trades?
No. The skill says the goal runtime is for research only and must never be used to place, submit or execute trades. It refuses or reframes live trading instructions as research-only analysis.
Why does the agent say my goal is blocked?
The agent uses blocked or insufficient_evidence when evidence is missing, stale, contradictory or cannot be verified. Completion needs verified evidence from an existing run_id or an allowed artifact_path with a matching sha256 hash.
Does the skill run scripts or use the internet?
The listing says it runs no scripts and uses no network. Any data lookups come from the agent runtime's tools, not from this skill file.
When should I not use research-goal?
Skip it for quick one-shot finance questions, since the skill itself says not to open a goal for those unless you ask. It is also not a source of investment advice.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use research-goal if you run multi-step finance research on the Vibe-Trading platform and want each conclusion tied to recorded evidence and open criteria. Pick something else if you need live trade execution, quick one-off answers, or a skill that works without the Vibe-Trading runtime.
Download the skill