Shocked by the Plunge in Gold Prices? I Used Kimi K2.5 + Cherry Studio to Create a “Review Tool” (Includes Agent Design + Full Tutorial)

Recently gold plunged, and many people's first reaction was: should I exit? Should I buy the dip? But looking back, the asset class gold is best at is “giving the market intensity.” Its violent swings often show echoes of history:

• Macro expectations suddenly shift(interest rates/inflation/stronger dollar), gold can quickly pull back

• Risk events heat up(conflict, financial system stress), safe-haven demand will push prices up

• Liquidity tightensand you can even see the counterintuitive pattern of “fall first, rise later”

The problem is: scrolling through ten news items produces emotion; what you need isa chain of evidence. Coincidentally, recently Moon's Dark Side released and open-sourced Kimi K2.5 model. — It is Kimi's smartest and most versatile open-source model to date, achieving open-source SOTA on Agent, code, image/video and other tasks.

So I had a bold idea: since the human brain can't handle so much messy information,could we put Kimi K2.5 inside an Agent in Cherry Studio to help me dig into this “gold crash” thoroughly?

Today's piece is not a dry manual, but will take you to use the latest model and hardcore Agent skills to outfit yourself with a 24/7 financial analysis team. At the end I'll provide the folder for this Agent Kimi Agent.

After you download and configure it for 3 minutes you can run it. Below I’ll expose the design logic, folder structure, component breakdown, and run process. If you stare at candlesticks worrying “should I buy the dip,” or are bombarded by news but can't find the real cause, you need this.

Why design it this way? (3 hard logics, no detours)

1. Data truth first, zero tolerance for fabrication: Financial analysis fears “AI hallucinations” most. So we force every step to annotate source + timestamp, and throw an error if unavailable (no guessing). Public sources (like Kitco, Investing.com) ensure zero API key barrier.

2. Modularize tasks + run in parallel: Kimi K2.5's highlight is the “Agent cluster” (autonomous clones, parallel 1500 steps). We simulate with Cherry Studio's Skills + Sub-agents: data fetching, news gathering, and report generation run in three streams, doubling efficiency.

3. Modernize outputs: No plain Markdown (who reads plain text anymore?), directly generate HTML (Chart.js charts + responsive layout), leveraging Kimi K2.5's code-generation ability.

Result: one report = near-year price chart + crash timeline + three-scenario forecasts + full source links. Send to your boss or group, ready to use.

📁 Folder structure: why compatible with Claude Code?

The core is .claude/ directory — Cherry Studio recognizes this and can auto-load Skills and configuration. The complete structure comes from your 06-DIRECTORY_STRUCTURE.md:

Kimi Agent/                           # project root directory
├── .claude/                          # Cherry Studio Agent core configuration area
│   ├── prompts/                      # system prompts (bilingual)
│   │   ├── system_prompt_cn.md       # Chinese version: defines Agent behavior + data truth rules
│   │   └── system_prompt_en.md       # English version
│   ├── skills/                       # 3 core Skills (auto-detected)
│   │   ├── skill_financial_data_fetcher.md     # data fetching + validation
│   │   ├── skill_geopolitical_analyst.md       # event analysis + timeline
│   │   └── skill_financial_report_generator.md # HTML report generation
│   ├── agents/                       # Sub-agents configuration
│   │   ├── subagent_financial_intelligence.md  # quantitative analysis submodule
│   │   └── subagents_usage_strategy.md         # collaboration strategy
│   ├── config/                       # paths/tools configuration
│   │   └── paths.conf
│   ├── settings.json                 # main config: model + prompt paths + tool list
│   └── mcp.json                      # MCP (tool server) config
├── docs/                             # documentation backup (README, data sources, changelog)
├── start-gold-agent.sh               # one-click start script (optional)
├── USER_PROMPT_EXAMPLE.md            # example instructions
└── backup directory (core_config/ skills/ etc.)# original file backups, not used at runtime

Why split like this?

• .claude/ It's Cherry Studio's standard recognition path: select the working directory and it auto-loads Skills (file names skill_*.md → Skill name financial-data-fetcher).

• Backup area prevents loss: original Skills are in the root skills/, at runtime use .claude/skills/.

🔧 Core component breakdown: 3 Skills + plugins + Sub-agents

🧩 Details of the three core Skills' design

Let's look at how these three “clones” are specifically designed and why.

Skill A:financial-data-fetcher (data hunter) — refuse hallucinations

• Design pain point: General LLMs most easily “make up” prices. Ask about gold prices and it might invent 2023 data.

• Skill logic:

  • Hard constraints: We wrote rules into the Prompt —“Forbidden to use prices from training data, must call tools”.

  • Toolchain: equipped with WebFetch. It doesn't “search” Baidu but directly “scrapes” specified data source pages (e.g., Kitco, GoldPrice.org, LBMA).

  • Data cleaning: it will clean the messy scraped HTML into clean JSON format (timestamp, open, close, change).

• Kimi K2.5's role: leverage its powerfullong-document extraction ability, to pinpoint the number like $2,xxx.xx from tens of thousands of lines of web code.

Skill B:geopolitical-analyst (geopolitical logic library) — refuse noise

• Design pain point: Many causes can explain a gold crash (Dollar up? War? Dumping?). Ordinary search will also pull in fake news from marketing accounts.

• Skill logic:

  • Cross-source verification: not only search “gold price,” but simultaneously search “Dollar Index (DXY),” “Fed minutes,” and “geopolitical situation.”

  • Time alignment: it will execute a core logic —“TimeStamp Matching”.

    • Finds: Gold crashed at UTC 14:30.

    • Search: What happened at UTC 14:30?

    • Match: Finds the US released a surprise CPI at UTC 14:30.

    • Conclusion: The crash was triggered by inflation data.

• Kimi K2.5's role: Using its Agent cluster (clones) capability, it can simulate “reading 20 news articles simultaneously” and filter out emotional noise, leaving only facts.

Skill C:financial-report-generator (front-end engineer) — refuse mediocrity

• Design pain point: This is also Cherry Studio's most impressive step. Most Agents will spit out a block of Markdown text, even tables are wonky.

• Skill logic:

  • Code-first: This Skill is trained to “speak only code.” It doesn't write articles; it writes HTML + CSS + JavaScript.

  • Dynamic interaction: Even if you lack programming basics, this component calls Chart.js library to turn data fetched by component A into a scalable, hoverable candlestick chart.

  • Visual integration: it will embed component B's analysis conclusions into the webpage layout as “cards” or a “timeline.”

• Kimi K2.5's role: Leveraging its upgraded Code (programming) ability, especially front-end building. Kimi K2.5 generates highly robust code that usually runs in the browser with minimal manual debugging.

What is the role of Sub-agents in this Agent? 🧩

Below I'll clarify the “Sub-agent” layer in the Kimi Agent (Gold Market Analysis Agent) : what they are, why to use them, how they collaborate, and what you'll see in the folder.

To be clear: Skills are more like “reusable process modules”;Sub-agent is more like a “dedicated role with its own work manual.”

The main Agent (the Gold Market Analysis Agentyou create in Cherry Studio) is responsible for three things:

1. Task decomposition: split “analyze gold trend/review crash/write report” into several independent subtasks

2. Task assignment: dispatch subtasks to different Sub-agents (each with clear boundaries and output format)

3. Acceptance and aggregation: check whether data has sources and timestamps, whether gaps exist; finally hand to the report generator to produce HTML

Why not let one Agent do everything?

• Because “fetch data, read news, compute indicators, write front-end report” require different context and tool calls; mixing into one Prompt easily goes off-track.

• After splitting, each sub-agent's rules can be written stricter:which tools are allowed, what output structure, how to handle failures.

Which Sub-agents are included in this configuration? What do they do? ✅This package's Sub-agents are mainly three types (two system preset, one custom):

A. System preset:search-specialist(search and information organization)

• name: search-specialist

• Responsibilities: advanced search, filter results, cross-source verification, organize citations

• Output characteristics: will provide search strategy, source URLs, key quotes (suitable for making a “crash trigger timeline”)

In gold analysis it typically handles:

• news sources, publish times, key sentences related to the “crash”

• official/authoritative source pages for central bank and macro data releases (e.g., CPI, rate decisions)

• multi-source validation of the same indicator (e.g., Kitco vs GoldPrice vs Investing)

B. System preset:business-analyst(indicator and correlation analysis)

Its tools are Read, Write, Bash, well-suited forstructured analysis:

• correlations (gold vs DXY, gold vs real rates)

• ETF position changes (e.g., SPDR Gold Trust)

• KPI calculations (annualized volatility, drawdown, etc. — assuming real data is obtained)

Its value is:turning “descriptions that look like analysis” into “computable conclusions with intermediate steps.”

C. Custom Sub-agent:financial-intelligence-agent(historical data/technical indicators/forecasting)

Path:.claude/agents/subagent_financial_intelligence.mdIt covers more “quant pipeline” work:

• pull historical data (OHLCV, economic indicators, rates, inflation, etc.)

• compute RSI / MACD / Bollinger Bands / moving averages / volatility

• output a set of traceable intermediate files: CSV, JSON (for example gold_technical_indicators.csv,correlation_analysis.json,gold_price_forecast_12m.csv)

This layer is especially critical: it strips “technical analysis” out of chat content and turns it into tangible deliverables. Then reuse, comparison, and sharing become convenient.

How are Sub-agents scheduled? (parallel strategy) ⚙️

This system prioritizes parallelbecause gold post-mortems are naturally multi-source information tasks.

Phase 1: parallel collection (reduce waiting)

• search-specialist: search “key news/data release timeline on the crash day”

• financial-intelligence-agent: pull near-one-year price series + calculate indicators

• business-analyst: compute correlations, organize ETF/macro explanatory framework

Phase 2: serial computations (dependencies later)

• Only after historical data is written to disk do we compute indicators/volatility/supports and resistances

• If data gaps are found, return to search-specialist source completion

Phase 3: aggregation and delivery

• The main Agent “aligns timestamps and metrics” of the three streams' results

• then calls the report generator to output HTML (with charts, timeline, citation list)

This is why this Agent performs better in “hot scenarios”: A gold crash = information-dense + inconsistent metrics, parallel collection + verification + aggregation can significantly reduce the “read a lot but are more confused” situation.

---

Sub-agent and Skills relationship: don't confuse them 🤝

In your package, the two are complementary:

• Sub-agentSub-agents

• : more like “specialized work modes” that solve “who does it, how to do it, which tools to use, what format to output.”Skills (.claude/skills/)

  • financial-data-fetcher: more like “reusable callable process modules” that solve “how to reliably execute this step.” For example:

  • geopolitical-analyst: emphasize multi-source verification, forbid fabricating numbers, output structured data

  • financial-report-generator: emphasize event classification, causal mechanisms, timeline format

: emphasize HTML templates, Chart.js, source lists, printable styles Simply put:

Sub-agents are responsible for distributing work; Skills ensure each step is more stable and reusable.

How do you confirm Sub-agents are “actually working” in the folder? 🔍

1. Check two places and that's enough:

   1. Whether directory is standard .claude/agents/ contains

2. subagent_financial_intelligence.mdRun logs

   1. (inside Cherry Studio)

   2. you will see traces of tool calls and task assignments (WebSearch/WebFetch/Bash/Write) Final deliverable files: for examplegold_analysis_report.html

, and if configured intermediate artifacts will also output CSV/JSON

If you want it more obvious, you can add a hard constraint in the main Prompt:

“Please list in the report appendix which sub-agents/skills were called this run and the filenames each produced.”

Then readers can tell at a glance: this is not chat, this is a pipeline.

🛠️ Practical tutorial: three steps to replicate your own AgentNo coding, no environment setup. I packaged a“Kimi Agent” folder

that you can use with just copy-paste. Step 1: model configuration (Moon's Dark Side's kimi-K2.5

+ Anthropic endpoint)

1. This is the core that makes the AI smarter. Open Cherry Studio → Model Services → click

1. Moon's Dark Side go to Moon's Dark Side open platform to obtainAPI Key

(required for model calls; data fetching does not require extra keys) ⚠️ High-energy alert (mandatory): In Cherry Studio's configuration, change “Endpoint Type” to.

• Anthropic Why change it?

Because Cherry Studio's Agent protocol needs to run in Anthropic endpoint mode so Kimi K2.5 can perfectly orchestrate the Skills mentioned above. Kimi Agent)

Step 2: create the Agent (directly mount the folder Kimi Agent I prepared for you at the end of the article named

1. which preinstalls all skills so you don't need to manually rewrite Skills/Sub-agents. + Cherry Studio → click next to the assistant list →

2. Create AgentGold Market Analysis AgentName:

3. (or any name you prefer). Model: choose the one you just configured, Step 1: model configuration (Moon's Dark Side's

4. Moon's Dark Side / Kimi Agent Working directory: choose the

1. folder you downloaded and extracted Open the system prompt in the folder (for example.claude/prompts/system_prompt_cn.md ) and paste it into Cherry Studio'sSystem Prompt

box.

1. Step 3: enable tool permissions + plugins + Skills (check all)Enable permissions

• : in the Agent config enable permissions and authorize tools (missing any may block):

• bash

• fetch

• edit

• multiedit

• webfetch

• web search

1. write

Configure plugins:

• business-analyst

• search-specialist

add system preset plugin:

• system skills:

Excel Analysis

Witness the “magic” Everything ready. Open the folder'sUSER_PROMPT_EXAMPLE.mdwhich contains a prewrittendeep instruction— copy and send it to the Agent.

1. This instruction will make the Agent do three things:Search

2. : search for the exact drop magnitude and time of the gold crash.Find

3. : using Kimi K2.5's web capability, locate Moon's Dark Side official descriptions of the new model's features (no fabrications allowed).Compare

: find historically similar crash patterns and analyze whether there is any “analogy.”

📊 Final presentation: what does it deliver?Click send and you'll see the Agent start running frantically. Logs will show: -> Thinking... -> Searching News...Calculating... After a while, you won't get nonsense; you'll receive an:

• HTML-formatted deep research report 📈Near-one-year gold price visualization

• (charts + tables) 🧷Key surge/crash interval annotations

• (especially the “crash”) 🗓️Event timeline

• (each event with source link for verification) 📊Technical indicators/correlation analysis

• (calculated when data exists; if missing, explicitly stated) 🔮Three-scenario forecasts

• (conditions, ranges, and risk warnings clearly written) 🧾Data source list

(URLs + fetch timestamps) In the end, you'll get a gold_analysis.html

---

file that opens instantly:

💭 Final notes What surprised me most in this experience was not how powerful Kimi K2.5 became, nor how easy Cherry Studio is to use. It was“certainty”

. In financial markets, information is money.We used to rely on guesses; now we rely on Agents. By putting Kimi K2.5 into Cherry Studio, we essentially hired ourselves a“absolutely rational, 24/7 connected, data-traceable”

super employee. You may not have avoided this gold crash, but if you learn this Agent workflow, at least on the cognitive dimension you've already earned it back.When you master“decomposing complex tasks into Skills components” Kimi K2.5 this core logic, plus the qualitative change in task planning and tool invocation, you'll find tasks you once thought “AI couldn't do” can now be handed to Agents:

• 🕵️♂️ Market scout: Don't want to manually comb competitor sites? Let an Agent automatically fetch the latest prices and feature updates from 10 competitors, clean and dedupe, and push a compiled Excel comparison to your desktop at 9 AM every day.

• 💻 Shadow programmer: Can't finish code? Beyond completion, let an Agent read the entire project folder, automatically write feature modules per requirements, run local tests, fix bugs, and generate a perfect API doc.

• ✈️ Ultimate traveler: Refuse bland itinerary notes. Let an Agent compare flights and hotels in real time against your budget, combine weather and local event reviews, plan a minute-precise itinerary, and even generate a PDF travel book.

The real charm of Agents isn't how long they can chat with you, but their autonomy and delivery capability— they can, like today's gold analyst, quietly finish the work while you drink coffee. Once you experience this“task automation”you can't go back.

We sincerely invite you to jump out of the box and explore more hardcore, fun, and practical scenarios. Whether workflow optimization or life hacks, please share your ideas and Agent config files.

📩 Contributions & communication:[email protected] Don't wait for the future. Your AI Agent era has already begun from this moment.

👇 Download now, plug in Kimi K2.5, and build your first digital team:

📥 Appendix: Kimi Agent configuration folder download link:https://pan.quark.cn/s/1ef986d1a9ff(Please ensure Cherry Studio v1.7.0+ is installed)