Stock Research Desk

Claude Code skill for multi-agent equity research. Produces buy-side memos with debate, scenario projection, and bilingual DOCX delivery. Use when researchin...

Security Scan

The skill's description promises a 12‑agent Python CLI pipeline, but the package provides only an instruction file (no code) while instructing you to install and run a local CLI—this mismatch and the broad file/web tools requested are incoherent and warrant caution.

Purpose & Capability

The name/description describe a complex 12-agent Python tool and CLI (./bin/research-stock). The skill declares python3, pip and an OLLAMA_API_KEY which would be reasonable for such a tool, but there are no code files, no install spec, and nothing in the bundle that implements the described pipeline. Asking the user to pip install -e .[dev] and run a local binary is inconsistent with the instruction-only packaging.

Instruction Scope

SKILL.md instructs installing a local package, setting OLLAMA_API_KEY in a .env, calling ./bin/research-stock and producing DOCX on the desktop, and describes web scraping/filtering behavior. Those runtime commands reference local files/binaries that are not present in the skill bundle. The instructions also allow web fetch/search and file read/write/edit operations — appropriate for research, but the combination with missing code means the agent could be given broad discretion without implemented constraints.

Install Mechanism

There is no formal install spec in the skill bundle (lowest risk), but SKILL.md tells users to run `pip install -e .[dev]` which presumes a local Python package. Because no package files are included, that instruction is misleading; it does not create an external download risk, but it is incoherent and could lead users to fetch the referenced GitHub repo or run arbitrary pip installs themselves.

Credentials

Only OLLAMA_API_KEY is required and declared as the primary credential, which is proportionate for an LLM-backed research tool. No unrelated secrets or multiple credentials are requested.

Persistence & Privilege

always:false and no requested config paths. The skill permits read/write/edit and web tools which are reasonable for a research/CLI tool, but these capabilities combined with missing code increase the need for caution.

What to consider before installing

This skill claims to be a complex Python CLI but contains only an instruction file — there is no code to implement the 12-agent pipeline. Before installing or running anything: (1) verify the upstream repository (the homepage) and confirm the package files and a safe install procedure exist; (2) do not paste your OLLAMA_API_KEY into unknown tools — prefer a scoped/limited key or test in a sandbox; (3) be cautious if following the pip install advice — it may require pulling code from the GitHub repo; (4) if you want to proceed, ask the publisher for the code or a packaged release, review the source, and run it in an isolated environment. Because of the mismatch between claims and provided files, treat this as suspicious until you can inspect the actual implementation.

A 12-agent multi-factor equity research desk for single-name deep dives, theme screening, recurring watchlists, and bilingual (Chinese + English) document delivery.

Quick Start

# Deep dive a single stock
research-stock 赛腾股份 中国

# Screen a theme
research-stock screen "中国机器人" --market CN --count 3

# Watchlist
research-stock watchlist add 赛腾股份 --market 中国 --interval 7d
research-stock watchlist run-due

The 12-Agent Pipeline

Step	Agent	Search?	Focus
1	market_analyst	Yes	Macro cycle, industry structure, China narrative
2	macro_policy_strategist	Yes	Interest rates, credit cycle, policy transmission
3	company_analyst	Yes	Business quality, management, financials
4	catalyst_event_tracker	Yes	Earnings dates, insider activity, M&A, regulatory
5	sentiment_simulator	Yes	Narrative temperature, participant psychology
6	technical_flow_analyst	Yes	Price action, volume, institutional flow, options
7	comparison_analyst	Yes	Peer comparison, relative valuation anchors
8	quant_factor_analyst	Yes	Factor exposure, statistical significance, regime
9	committee_red_team	No	Contrarian challenge, hidden fragility
10	guru_council	No	Multi-perspective synthesis (Buffett/Druckenmiller/Simons)
11	mirofish_scenario_engine	No	Bull/base/bear scenario projection
12	price_committee	Yes	Target prices with explicit horizons

Workflow

1. Install: pip install -e .[dev]
2. Set OLLAMA_API_KEY in .env
3. Run ./bin/research-stock <name> <market> for a full memo
4. Or ./bin/research-stock screen "<theme>" --market CN --count N for screening
5. Output: bilingual DOCX on desktop with Chinese section first, English section second

Source Quality Model

Domain-level source scoring filters out noise. Official filings (cninfo.com.cn, SEC) score 94-96. Quality media (yicai, caixin) score 84. Forum noise (<36) is blocked entirely.

Safety

• No trading execution
• No portfolio management
• No backtesting engine
• Target prices always tied to explicit horizons and theses