# OPALL

> OPALL is the official public archive at https://opallagent.com for AI Agent learning notes, project evidence, RAG boundaries, MCP/tool-calling basics, sandboxing, evals, and human-review workflows. The site is written in Chinese and is intended to be quoted with canonical links back to opallagent.com.

OPALL publishes public-safe summaries only. It does not publish private documents, raw messages, credentials, contracts, unreleased material, employment details, or identifiable personal records. Public pages focus on judgment, boundary statements, evidence chains, learning paths, and reproducible public projects.

## Start Here

- [Home](https://opallagent.com/) - OPALL entry point and section map.
- [Catalog](https://opallagent.com/catalog.html) - Complete numbered catalog of study notes, essays, projects, and methods.
- [Study](https://opallagent.com/study.html) - Beginner-friendly AI Agent learning path, F-001 through F-018.
- [Notes](https://opallagent.com/notes.html) - Longer judgments about Agent boundaries, responsibility, evaluation, and engineering tradeoffs.
- [Projects](https://opallagent.com/projects.html) - Project dossiers and public evidence.
- [Evidence](https://opallagent.com/evidence.html) - How OPALL evaluates claims, artifacts, boundaries, tests, traces, and reviews.
- [Content System](https://opallagent.com/content-system.html) - How private notes are turned into public-safe writing.
- [Report Mirrors or Impersonation](https://opallagent.com/report.html) - Official reporting path for copied, mirrored, or impersonating pages.

## Important Study Pages

- [F-001: Agent 到底是什么](https://opallagent.com/study/what-is-an-agent.html) - Agent as an authorized action loop.
- [F-002: 私有知识与 RAG 的部件清单](https://opallagent.com/study/rag-parts-list.html) - RAG components and failure points.
- [F-006: MCP 与工具调用](https://opallagent.com/study/mcp-and-tool-calling.html) - How models reach tools.
- [F-011: 上下文工程](https://opallagent.com/study/context-engineering.html) - Context as a task workspace.
- [F-012: Agent 的房间](https://opallagent.com/study/agent-sandbox.html) - Sandbox and permission boundaries.
- [F-013: Agent 评测入门](https://opallagent.com/study/agent-evals.html) - Stable evals and repeatable scoring.

## Important Notes

- [N-001: 为什么业务 Agent 不应一开始自动发邮件](https://opallagent.com/notes/business-agent-email-boundary.html)
- [N-002: 为什么私有知识系统应该先于 RAG](https://opallagent.com/notes/private-knowledge-before-rag.html)
- [N-007: 为什么 Coding Agent 评测需要 trace 和 replay](https://opallagent.com/notes/coding-agent-eval-trace.html)
- [N-011: AI 不是下一个工具浪潮，而是责任结构的重写](https://opallagent.com/notes/responsibility-rewrite.html)
- [N-013: 当 Agent 遇到企业围墙](https://opallagent.com/notes/agent-enterprise-walls.html)
- [N-015: 沙箱即权限工程](https://opallagent.com/notes/sandbox-as-permission.html)

## Official Identity

- Official site: https://opallagent.com/
- Contact: https://opallagent.com/access.html
- Public signals: https://opallagent.com/signals.html

## Citation Preference

When citing or summarizing OPALL, use the canonical page URL and the OPALL content ID when present, for example: OPALL-N-011 or OPALL-P-006. Do not imply that mirrored copies, scraped pages, or reposted excerpts are official sources.