docs(education): add eval-driven-development examples page

docs/education/README.md

@@ -0,0 +1,98 @@
+<!-- START doctoc generated TOC please keep comment here to allow auto update -->
+<!-- DON'T EDIT THIS SECTION, INSTEAD RE-RUN doctoc TO UPDATE -->
+**Table of Contents**  *generated with [DocToc](https://github.com/thlorenz/doctoc)*
+
+- [Maintainer education stream](#maintainer-education-stream)
+  - [Who this is for](#who-this-is-for)
+  - [What the stream covers](#what-the-stream-covers)
+  - [Framework posture this stream teaches](#framework-posture-this-stream-teaches)
+  - [Relationship to other docs](#relationship-to-other-docs)
+  - [Placeholders and project-agnosticism](#placeholders-and-project-agnosticism)
+  - [Licence](#licence)
+
+<!-- END doctoc generated TOC please keep comment here to allow auto update -->
+
+<!-- SPDX-License-Identifier: Apache-2.0
+     https://www.apache.org/licenses/LICENSE-2.0 -->
+
+# Maintainer education stream
+
+Building an agentic project is a different craft from writing a service or a
+CLI. Behaviour is **probabilistic, not deterministic**. Prompts and skill files
+**are code** in every meaningful sense — they are diffed, reviewed, and
+upstreamed like code. Evaluating an agent's output is **harder than testing a
+function**. The unit of authorship shifts from "a function in a file" to "a
+skill the agent invokes".
+
+This stream exists because a platform no one knows how to use safely is not
+adoptable, regardless of code quality (PRINCIPLE 18). Every Magpie release
+ships the education material maintainers actually need for the skills that
+release includes.
+
+## Who this is for
+
+- **Maintainers adopting Magpie** for the first time and wondering where to
+  start.
+- **Maintainers who have adopted Magpie** and want to extend it with their own
+  skills or adapt existing ones for their project.
+- **Contributors to the Magpie framework** who want to understand the mental
+  models behind the design choices.
+
+If you are evaluating whether to adopt at all, start with
+[MISSION.md](../../MISSION.md) and [PRINCIPLES.md](../../PRINCIPLES.md) first
+— the education stream assumes you have already decided to engage.
+
+## What the stream covers
+
+| Page | What it teaches | Status |
+|---|---|---|
+| **This page** | What the stream is and how to navigate it | Stable |
+| `pattern-catalogue.md` | Copy-pasteable skill / prompt / tool-use patterns with war stories | Planned |
+| `your-first-skill.md` | Zero-to-merged path for landing a first working skill | Planned |
+| [`eval-driven-development.md`](eval-driven-development.md) | How to think about correctness when "correct" is a distribution | Stable |
+| `workshops.md` | Office-hours format, scheduling, recordings | Planned |
+
+Pages marked **Planned** are scheduled in the implementation plan and will
+land in upcoming releases. Each planned page will add its own row as a live
+link when it arrives — the index stays link-clean before then.
+
+## Framework posture this stream teaches
+
+Every example in this stream inherits the framework's baseline posture — not
+as a disclaimer, but as worked demonstration:
+
+- **Data, not instructions** (PRINCIPLE 0) — external content (issue bodies,
+  PR comments, mail) is treated as data routed through the Privacy-LLM gate or
+  redacted before reaching a model. Skills never pass external text as
+  system-level instructions.
+- **Privacy and sandbox by default** (PRINCIPLE 1) — agents run in a locked
+  sandbox; skills declare their tool permissions explicitly.
+- **Eval as release discipline** (PRINCIPLE 8) — every skill ships an eval
+  suite; examples in this stream use the in-repo harness
+  (`tools/skill-evals/`) rather than an ad-hoc approach.
+
+## Relationship to other docs
+
+- **[magpie-write-skill](../../.claude/skills/magpie-write-skill/SKILL.md)**
+  is the authoring *reference* for someone who already knows the shape of a
+  skill. The `your-first-skill.md` page (planned) is the *beginner on-ramp*
+  that gets you to the point where the reference is useful.
+- **[tools/privacy-llm/pii.md](../../tools/privacy-llm/pii.md)** is the PII
+  redaction reference catalogue. The pattern catalogue (planned) is a teaching
+  artefact — it shows *how* and *why*, not just *what*.
+- **[docs/rfcs/RFC-AI-0004.md](../rfcs/RFC-AI-0004.md)** established the
+  principles this stream operationalises. The RFC's forward reference to the
+  "maintainer-education stream" resolves here via
+  [MISSION.md](../../MISSION.md).
+
+## Placeholders and project-agnosticism
+
+All examples use `<PROJECT>`, `<tracker>`, `<upstream>`, and
+`<security-list>` placeholders (PRINCIPLE 12). A concrete adopter name in a
+skill or a pattern is a refactor bug; swap your config, not the material.
+
+## Licence
+
+Content in `docs/education/` is Apache License 2.0 (PRINCIPLE 17).
+AI-authored contributions carry a `Generated-by:` token in the commit message,
+per ASF Generative Tooling Guidance.