UZI-Skill is a collection of coding-agent skills for analyzing individual stocks across Chinese, Hong Kong, and United States markets using public data, investor viewpoints, quantitative rules, and institutional analysis methods. It is for users who want an agent to produce detailed stock-analysis reports. The catalogue entries are the commands, skills, instructions, plugin, agent, and hook that provide this workflow in supported coding agents.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/wbh604/UZI-SkillWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/commands/wbh604/uzi-skill/analyze-stock)<a href="https://agentmods.dev/commands/wbh604/uzi-skill/analyze-stock"><img src="https://agentmods.dev/badge/commands/wbh604/uzi-skill/analyze-stock/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/commands/wbh604/uzi-skill/analyze-stock"><img src="https://agentmods.dev/badge/commands/wbh604/uzi-skill/analyze-stock.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00046 | $0.02329 |
| Opus 5 | $0.00023 | $0.01164 |
| Sonnet 5 | $0.00009 | $0.00466 |
| Haiku 4.5 | $0.00005 | $0.00233 |
Grade A, and why
analyze-stock scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 170 lines — stays where its author put it; the contents beside it link to each section on GitHub.
深度分析任务
用户输入: $ARGUMENTS
执行流程(两段式 · 你必须在中间介入)
第一段 · 数据采集 + 骨架分(脚本完成)
cd <plugin_root>
pip install -r requirements.txt 2>/dev/null
cd skills/deep-analysis/scripts
python -c "from run_real_test import stage1; stage1('$ARGUMENTS')"
这会跑完 Task 1 → 1.5 → 2 → 3(规则引擎骨架分),输出到 .cache/{ticker}/ 下。
第二段 · 你来分析(核心!不能跳过!)
Stage 1 跑完后,你必须做以下事情:
0. v2.13.5 · Playwright 兜底前置(必走)
import json, os
from pathlib import Path
net = json.loads(Path(".cache/_global/network_profile.json").read_text(encoding="utf-8"))
issues = json.loads(Path(f".cache/{ticker}/_review_issues.json").read_text(encoding="utf-8"))
low_quality_dims = [
i["dim"] for i in issues.get("issues", [])
if i.get("category") == "data" and i.get("severity") in ("critical", "warning")
]
if low_quality_dims:
os.environ["UZI_PLAYWRIGHT_FORCE"] = "1"
from lib.playwright_fallback import autofill_via_playwright
autofill_via_playwright(raw, ticker) # 主动强制再跑一次 · 补数据
1. 读取评委骨架分
读 .cache/{ticker}/panel.json,看 65 人各自打了多少分。特别关注:
- Top 5 看多和 Top 5 看空分别是谁?他们的 headline 有没有说服力?
- 有多少人 skip 了?(非 A 股时游资会 skip)
- 有没有明显不合理的分数?
2. 逐组分析(spawn 4 个并行 sub-agent)
对每组投资者,spawn 一个 Agent:
Agent 1 · 价值 + 成长派(10 人)
你要扮演巴菲特/格雷厄姆/费雪/芒格/邓普顿/卡拉曼/林奇/欧奈尔/蒂尔/木头姐,
逐一对 {stock_name} ({ticker}) 给出判断。
公司数据:{从 raw_data.json 摘取关键数据}
规则引擎参考分:{从 panel.json 摘取这 10 人的 score/headline}
真实持仓:{巴菲特持有苹果/BYD, 段永平持有苹果/茅台/腾讯 等}
对每人输出: investor_id, signal, score(0-100), headline(引用数字), reasoning(2-3句)
你可以覆盖规则引擎的分数——你是在模拟这个人的判断,不是跑公式。
Agent 2 · 宏观 + 技术派(9 人) Agent 3 · 中国价投 + 量化(9 人) Agent 4 · 游资(23 人) — 非 A 股直接全部 skip
3. 合并 agent 结果
把 4 个 agent 返回的 {signal, score, headline, reasoning} 覆盖到 .cache/{ticker}/panel.json 的对应投资者上。
4. 写 agent_analysis.json(闭环关键!)
对关键维度(财报/估值/护城河/行业)写 1-2 句定性评语(≥20 字,引用具体数字)。如果需要,web search 补充信息。
⚠️ 必读:agent_analysis.json 完整 schema(缺字段 stage2 会报 schema warning/error)
| 字段 | 要求 | 触发校验 |
|---|---|---|
agent_reviewed |
必须 true |
⚠️ 缺 → warning |
dim_commentary |
覆盖全部 22 维,每条 ≥20 字(引用具体数字,禁止空泛) | ⚠️ <20 字 → warning |
panel_insights |
≥30 字,评委投票分布 + 多空分歧分析 | ⚠️ <30 字 → warning |
great_divide_override |
punchline(≥10 字) + bull_say_rounds(≥3 条) + bear_say_rounds(≥3 条) | 🔴 缺字段 → error |
narrative_override.core_conclusion |
≥20 字综合定论 | ⚠️ <20 字 → warning |
narrative_override.risks |
≥3 条风险 | ⚠️ <3 条 → warning |
narrative_override.buy_zones |
必须含 value/growth/technical/youzi 四个 key,每个 key 内含 price(数值, youzi 可为 0) + rationale(≥5 字解释) |
🔴 缺 key → error / ⚠️ 缺子字段 → warning |
qualitative_deep_dive |
覆盖 3_macro/7_industry/8_materials/9_futures/13_policy/15_events 共 6 维。每维含:evidence 数组(≥2 条)、associations 跨域因果链(6 维合计 ≥3 条)、conclusion(1-2 句) |
🔴 evidence 非 list → error |
data_gap_acknowledged |
dict 格式 {"dim_key": "已尝试 X 但失败的原因"},标记数据采集失败但 agent 已知晓的维度 |
🔴 类型非 dict → error |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 170 lines · 46 tokens per session scan A c09c1ef155d0
analyze-stock is a command published in the GitHub repository wbh604/UZI-Skill (6,820 stars, last pushed 4d ago), licensed MIT. It adds 46 tokens to every session and 2,329 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other commands, from other repositories
valuation-methods
Valuation methods analysis — multiples, DCF inputs, PEG integration, valuation assumption extraction.
merit-reconcile
Preview or check the status of Stripe → Merit payout reconciliation (read-only).
audit-checklist
Perform an internal audit, review controls, or prepare for an external financial audit.
scan
Scan AWS account for cost optimization.
finops-status
Orientation — say where an opportunity or assignment sits in the five-step FinOps lifecycle and what unlocks next. Useful when a record has no active stage: an opportunity while its assignments do the work, an assignment whose plan has not been approved yet, or a rejected or archived assignment. Read-only; mutates…
audit
Scan for non-kernel money math. Rebuild critical flows as JournalEntrys. Replay and prove. Complements /ledger-verify.