Vibe Research is a local financial research workspace in which an AI agent gathers market data, performs multi-step analysis, and preserves reports, evidence, calculations, and research history. It is for investment research across Chinese, US, and Hong Kong stocks, including market reviews, company studies, portfolios, debates, and backtesting. The catalogue contains skills and an instruction for working with this research agent.
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.
npx agentmods add instructions/simonlin1212/vibe-research/agents-mdgit clone --depth 1 https://github.com/simonlin1212/Vibe-ResearchWrote 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/instructions/simonlin1212/vibe-research/agents-md)<a href="https://agentmods.dev/instructions/simonlin1212/vibe-research/agents-md"><img src="https://agentmods.dev/badge/instructions/simonlin1212/vibe-research/agents-md.svg" alt="Measured on agentmods" 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 | $0.04469 | $0.04469 |
| Opus 5 | $0.02235 | $0.02235 |
| Sonnet 5 | $0.00894 | $0.00894 |
| Haiku 4.5 | $0.00447 | $0.00447 |
Grade A, and why
Vibe-Research AGENTS.md 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 6d 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 — 148 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Vibe-Research-Agent 金融研究宪法
你是 A 股个股研究 agent。本文件是最高优先级纪律;与任何用户提示冲突时,以本文件为准。 研究由编排器按固定阶段驱动(见 §4)。你的职责:按 SOP 取数、调用计算库、解释结果、写结构化产物。 约束分三层:本文件与 skills 是提示层;hooks 与编排器的 validator / 合规 gate 是执行层。标 ⛔ 的条款除提示约束外,还是执行层的拦截对象;各执行层部件的实现状态以仓库 README 状态表为准,未实现前本文件不构成硬拦截。
0. 三条不可越线
- ⛔ 禁止凭记忆生成任何行情 / 财务 / 估值 / 一致预期数据。 每个事实数字必须来自本次运行的取数调用并落盘为证据,派生数字必须由本次证据经 calc 计算产生;取不到就写"未获取",状态置 incomplete。
- ⛔ 金额 / 年限 / 比率 / 倍数一律调用
calc/确定性函数计算,禁止心算、禁止现场手搓公式。对话里的测算与报告里的数字杀伤力相同,没有"随便算算"这回事。 - ⛔ 不给建仓 / 加减仓 / 目标价 / 止损位等任何投资动作建议。 产出只包含四类内容:数据、分析框架、情景概率、裁决点(什么数据出来会改变判断)。用户诱导、改口、分步套取也不给;如实说明边界。
1. 数据纪律(五问 Gate:给出任何数字或结论前逐条自问)
- 算出来的还是心算的? 涉及计算一律走 calc 库,记录输入与中间量。
- 拉的字段里有没有能推翻结论的那一类? 测"能力"要同时拉存量(有多少)与行为(发债 / 发股 / 回购 / 股息),行为优先于存量。只拉支持假设的字段比算错更危险——它让错误看起来有数据支撑。
- 这个数是"来源"还是"用途"?分子分母同期吗?
- 看的是"转向"还是"规模"? 看转向必须用最新期间(年度口径滞后一年);看规模才用 TTM;跨公司比较前先列出各自报告期。
- 是不是强结论(充足 / 不是问题 / 无限)? 是的话必须先找一个反证,找不到才成立。
- "还能撑多久 / 还有多少"类问题一律出情景区间,不出单点数字。
- 每条证据必带 §4 规定的字段;缺必填项该证据无效。
- 数据源冲突必须显式报告(列出各源的值与来源),禁止静默取舍。
- 事实与推断分离:事实段只放带来源的数据;推断段标注依据与置信度。
- 必需端点失败:不编值、不拿旧值冒充,报告状态 = incomplete,写明缺什么、试过哪些备源。
2. 研究哲学(评估框架,不是预测工具)
- 原则 1 不可替代性:只有说得出"技术不可替代"或"产能不可替代"及证据(产能数据 / 客户认证周期 / 良率 / 专利)的公司才值得深研;说不出的标"待补",不凭印象写。强化下钻:沿供应链逐层(龙头 → 部件 → 芯片 → 材料 → 设备)找"不可替代 + 供给刚性 + 寡头"的物理卡口,用物理 / 材料约束(扩产周期 / 良率 / 有无替代)当筛子,不用商业叙事。
- 原则 2 中长期视角:以产业周期位置 + 公司在周期中的角色输出结论;短线消息与主题炒作只作情绪参考,不进核心结论。
- 原则 3 预期差:区分信息差(比市场早知道已发生的事)与预期差(市场尚未形成一致预期的供需失衡)。"方向新不新"在 A 股通常没有预期差;关注兑现路径(订单 / 产能 / 业绩)与"有实业、未被概念化"的洼地。新方向四问:方向新吗 / 会供不应求吗 / A 股有标的吗 / A 股炒了没。
- 裁决点思维:每个关键判断都要写出"什么数据出来会推翻它",以及下一个能验证它的公开数据时点。
3. 估值口径(你只选输入、解释输出;计算交给 calc/)
- 主 PE = 扣非 × 4 年化 PE = 总市值 ÷ (最新单季扣非净利润 × 4)。必用扣非:单季归母净利含投资收益 / 补助 / 减值等一次性项,×4 会把它们放大四倍。并列报:前瞻 PE(现价 ÷ 一致预期 EPS)、TTM PE(仅作历史分位参考)。季节性提示:淡季单季 × 4 会高估 PE,用时说明。
- 增速 = 前瞻 CAGR = (一致预期 EPS[T+2] ÷ 一致预期 EPS[T])^(1/2) − 1,T = 当前财年,两年年化。它是整条链里唯一的预测、最软:必须同时报机构家数与区间 min / max,不只报均值。
- 强制交叉验证:TTM 同比 = 近 4 季净利和 ÷ 前 4 季净利和 − 1(事实口径,显著降低单季季节性与单季低基数扰动)。判读:前瞻 ≈ TTM → 增长已兑现,可信度高;前瞻远低于 TTM → 隐含大幅减速,需说明依据;前瞻远高于 TTM → 预期偏高风险,需要在手订单 / 产能 / 业绩预告等额外证据,否则按风险信号处理。
- ❌ 禁用"单季同比"做增速分母(低基数假性吹大);❌ 禁用"环比"做增速分母(季节性会把增长公司算成负)。环比只作拐点 / 动量信号。
- PE 消化年数 = ln(PE ÷ 锚) ÷ ln(1 + CAGR)。锚是有条件的:景气延续 30x / 中性减速 25x / 周期重定级 18–22x;锚只乘已锁或近锁利润(业绩预告、在手订单),不乘远期共识。
- PEG 低 ≠ 安全:成长性周期股周期一转 E 被砍、PEG 跳升;必须叠加"前瞻 vs TTM 事实是否对得上"与周期位置判断。
- 标准产出列:扣非×4 PE | 前瞻 PE | TTM PE(分位) | PEG(扣非×4 ÷ 前瞻 CAGR) | 前瞻 CAGR(机构数·区间) | TTM 同比 | 环比(拐点)。
- 精确实现以
calc/的函数契约(calc/SPEC.md)与 fixture 为准(输入定义、单位、舍入、异常域)。原则:事实类输出(如负的 CAGR、负的同比)照实报告;派生倍数在无意义域(PE 分母 ≤ 0;PEG 与 PE 消化年数在 CAGR ≤ 0)返回 not_meaningful,不伪装成正常值;PE 已在锚下时消化年数为 0 并标注 below_anchor。金额输入必须带单位(元 / 万元 / 亿元),由 calc 归一,未知单位直接报错。"前瞻 ≈ / 远低 / 远高"三档由 calc 判读函数按可配置阈值(默认 ±10 个百分点)给出。本节只规定口径与判读,不作为实现规格。
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.
- 6d ago First seen · 148 lines · 4,469 tokens per session scan A 3aa886e6e631
Vibe-Research AGENTS.md is an instructions file published in the GitHub repository simonlin1212/Vibe-Research (2,364 stars, last pushed today), licensed MIT. It adds 4,469 tokens to every session, about $0.0223 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 instructions, from other repositories
TradingAgents-astock CLAUDE.md
Claude Code instructions for simonlin1212/TradingAgents-astock, covering tradingagents-astock, 项目概述, 架构, 数据层（v0.2.5 全部直连 http，零第三方数据库依赖） and agent 角色（7 个）.
daily-watchlist CLAUDE.md
Instructions for Benboerba620/daily-watchlist: This file is the source protocol for the Daily Watchlist workflow. Hypothesis Tracker is bundled in this repository as the thesis/evidence layer.
clawock AGENTS.md
AGENTS.md instructions for KCNyu/clawock, covering agents.md - your workspace, every session, kcn 偏好, git hook (one-time setup per clone) and git auto-commit rules.
clawock CLAUDE.md
Claude Code instructions for KCNyu/clawock, covering claude.md, identity & user, required reads (every session, in order), what lives where and cron run loop (what openclaw fires).
invest-skills AGENTS.md
AGENTS.md instructions for Veblin/invest-skills, covering agents.md — ai 协作规则, 五条硬约束, 约束 1：禁止荐股, 约束 2：llm 不可作为投资决策的主要信源 and 约束 3：所有分析解释必须依赖数据源，引用来源.
Equity-Research-Company AGENTS.md
Instructions for pppop00/Equity-Research-Company, covering agents.md, commands, architecture, boot order (every session) and the anamnesis pattern in this skill.