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/agentpit-io/hunter-community/claude-mdgit clone --depth 1 https://github.com/agentpit-io/hunter-communityWrote 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/agentpit-io/hunter-community/claude-md)<a href="https://agentmods.dev/instructions/agentpit-io/hunter-community/claude-md"><img src="https://agentmods.dev/badge/instructions/agentpit-io/hunter-community/claude-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 | $0.01129 | $0.01129 |
| Opus 5 | $0.00564 | $0.00564 |
| Sonnet 5 | $0.00226 | $0.00226 |
| Haiku 4.5 | $0.00113 | $0.00113 |
Grade A, and why
hunter-community CLAUDE.md scanned grade A with 1 finding 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 5d 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.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
**前端验证不能只靠 HTTP 200 / `node --check` / curl** —— 这些都测不到内联 `<script>` 里的错。用 `apps/web/public/strategies/render_check.js`,在 node 的 `vm` 里真跑一遍页面脚本,假 `document` / `localStorage` / `fetch` 兜住。 How it starts
The opening of the file, as written. The whole thing — 65 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md — hunter-community
本文件是 Claude Code 在本仓库工作时会自动读取的项目指令。
铁律:严禁 mock 兜底 · 空的比假的好
量化 / 指标 / 因子 / 回测 / 评分 类代码,任何用户会看到或送给 LLM 分析的数字都必须是真算出来的。算不出就显示 — 并注明原因。
严禁的模式
Object.assign(mockDefault, realResult)— 真数据缺字段时假数据会露出realValue || 0/realValue ?? 0.5—0.00看起来像个结论- 渲染层直接写常量或伪造公式(基准
= 1 + i*0.005、超额= ann_ret - 0.062、相似度= [87,62,54][i]之类) - API 转换层挂
_mock/metrics_mock字段 — 修了展示层不算完,上游会继续注入 - 把不确定真假的指标送进
askHunter/ LLM — 模型会认真点评根本不存在的业绩
Why
2026-08-18 事故:「全能蓝筹」3 年回测同屏出现「年化 −1.2%」与「信息比率 0.98 / 月度胜率 62%」—— 后两个是 backtest.html 里写死的常量,因为后端 _calc_metrics() 从来没算过 ir / win_rate,Object.assign 用 mock 补上了。全量排查后共发现 4 个页面 8 处假数据(20 个因子的 60 个指标、6 个官方策略 metrics、净值曲线、基准直线、因子归因公式…),其中最糟的是 askHunter 把假指标发给 LLM 分析。
相关提交(按顺序读):
05c8d3d fix(quant): 回测不再用假数字冒充真结果 (_17 步1-3)
fefcb6c fix(quant): 清掉全部写死的假数据 (_17 全量排查)
13a6f8f feat(quant): 接真指数日线 + 实现回测基准
863d2d1 feat(quant): 指数成分股走代理拉取 + seed 真数据
ba62a96 fix(quant): 补完 _17 剩下四条
9335fe3 fix(quant): 修白屏 —— 漏了两处读取方与三类语法坑
演示模式不是借口 —— 要 demo 数据就跑一次真回测把结果存下来当样例,不要在渲染层编。
落地方法
改动前 —— 先全量 grep,把所有假数据来源列一遍再动手:
grep -rn "mock\|_mock\|Object.assign\|writeDefault\|\|\| 0" apps/web/public apps/api/app
修一处后要再全量排查一次 —— 上游 API 转换层常常在真数据上重新挂假字段。前科:先修了 backtest.html 的 mock,但 app.js:197 在 API 转换时又挂了 metrics_mock,真数据流下来被再次污染。
删/改字段名要 grep 所有读取方,不能只搜字段名本身:
- ❌ 只搜
metrics_mock - ✅ 还要搜
.metrics.、?.metrics、s.metrics等直接属性访问 - 前科:白屏事故就是漏了
s.metrics.ann_ret这种直接读取
前端验证不能只靠 HTTP 200 / node --check / curl —— 这些都测不到内联 <script> 里的错。用 apps/web/public/strategies/render_check.js,在 node 的 vm 里真跑一遍页面脚本,假 document / localStorage / fetch 兜住。
触发词
出现下列任一词的改动,自动触发本铁律:
回测 · 指标 · 因子 · 成分股 · 换手率 · 换仓频率 · 基准 · 超额 · IR · 信息比率 · 胜率 · 净值 · 归因 · 持仓 · 相似度 · 评分 · metrics · sharpe · ann_ret
详细文档
完整问题清单与实施记录(在 agentpit repo 内,不在本仓):
agentpit/doc/开源hunter-community/01详细工作目录/11量化策略/17_20260818_回测可信度问题与修复方案.mdagentpit/doc/开源hunter-community/01详细工作目录/11量化策略/18_20260818_回测可信度修复实施记录.md
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.
- 5d ago First seen · 65 lines · 1,129 tokens per session scan A d05dde333e44
hunter-community CLAUDE.md is an instructions file published in the GitHub repository agentpit-io/hunter-community (410 stars, last pushed yesterday), licensed Apache-2.0. It adds 1,129 tokens to every session, about $0.0056 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
plannotator AGENTS.md
AGENTS.md instructions for backnotprop/plannotator, covering plannotator, project structure, server runtimes, installation and environment variables.
solo AGENTS.md
AGENTS.md instructions for solo-agent/solo, covering project testing rules and project service lifecycle rules.
superset AGENTS.md
AGENTS.md instructions for superset-sh/superset, covering superset monorepo, project structure, exception: shadcn/ui components, database and releases.
agent-tools AGENTS.md
AGENTS.md instructions for kairyou/agent-tools, covering agent instructions, build and test, layout, settled decisions and docs.
untether GEMINI.md
Gemini CLI instructions for littlebearapps/untether, covering untether — gemini instructions, stack & conventions, architecture, key rules and testing.
superset CLAUDE.md
Claude Code instructions for superset-sh/superset, a project described as: Superset is an agentic IDE to orchestrate 100+ coding agents in parallel. Run any agent with your own subscription.