尚未检测到更新。GitHub 发布版本后会显示在这里。
个人 GitHub 账号下公开仓库的所有者可以登录并验证所有权,自行管理介绍,无需管理员批准。
登录并生成验证文件,在 30 分钟内将其提交到此仓库的默认分支。其他分支或分叉仓库中的文件无效。然后在下方检查文件,即可无需管理员批准直接管理项目。请保留文件以维持管理权限。
如需登录,将前往登录页面。登录后请回到此处,再次点击所选的验证按钮。
在项目文件夹中运行 /nomorevibe verify,即可验证所有权并自行管理介绍。
一分钟完成验证正在加载完整 README。 以下为已保存的摘录。
An evidence-first startup due diligence and investment intelligence platform. Give it a pitch deck, a website and notes; it extracts evidence, analyses eight dimensions, and produces a structured report with an Investment Score. Every statement is labelled as verified or sourced, AI analysis, assumption, or missing, and every claim can be traced to its source.
Status: pre-alpha. The specification set is complete; implementation starts at Phase 0 (see docs/TRACKER.md).查看完整 README ↗An evidence-first startup due diligence and investment intelligence platform. Give it a pitch deck, a website and notes; it extracts evidence, analyses eight dimensions, and produces a structured report with an Investment Score. Every statement is labelled as verified or sourced, AI analysis, assumption, or missing, and every claim can be traced to its source.
Status: pre-alpha. The specification set is complete; implementation starts at Phase 0 (see docs/TRACKER.md).
ARGUS AI is a research and intelligence tool. It does not provide investment, legal, tax or financial advice, and its outputs are not a substitute for professional due diligence. Verify all material facts independently.
pnpm install
cp .env.example .env.local # fill in the values
pnpm emulators # Auth, Firestore, Storage
pnpm dev # http://localhost:3000
Never commit .env.local or any real documents.
| Script | Purpose |
|---|---|
pnpm dev / pnpm build / pnpm start | Run, build, serve |
pnpm lint / pnpm format / pnpm typecheck | Static checks |
pnpm test / pnpm test:e2e | Unit and integration tests, Playwright |
pnpm check | Lint, typecheck and tests. Must pass before a task is done. |
pnpm emulators | Firebase emulators |
pnpm analyze <path> | Run the analysis engine on a fixture or folder |
pnpm eval | Run the eval harness (calls the live model) |
| Doc | Contents |
|---|---|
docs/PRD.md | Product requirements, epistemic model, report spec |
docs/TRD.md | Architecture, stack, API, security, budgets |
docs/AI_SPEC.md | Evidence contract, pipeline, scoring, verification, prompts, evals |
docs/SCHEMA.md | Zod schemas, Firestore model, rules, canonical fact keys |
docs/APP_FLOW.md | Routes, screens, states, edge cases |
docs/DESIGN.md | Design system |
docs/IMPLEMENTATION_PLAN.md | Phases and kickoff prompts |
docs/TRACKER.md | Task status |
docs/RULES.md | Engineering and AI rules |
docs/PROJECT_MEMORY.md | Decisions, conventions, gotchas, session notes |
CLAUDE.md | Entry point for Claude Code |
src/app routes and API handlers
src/components UI components
src/lib analysis engine, model client, schemas, repositories
evals/ fictional fixtures and eval harness
firebase/ security rules and indexes
docs/ specifications
Start each session from the repository root so CLAUDE.md loads. Use the kickoff prompt for the current phase in docs/IMPLEMENTATION_PLAN.md.
To be decided before any public release.