docs: add AGENTS.md (CLAUDE.md symlink) for AI coding agents (#20)
## What & why
Adds an **`AGENTS.md`** (with a **`CLAUDE.md` symlink**) at the repo
root — a concise operating manual for AI coding agents contributing to
this repository.
It is deliberately **minimal and non-duplicative**:
[`README.md`](./README.md) remains the source of truth for everything
mechanical (what a Server Card is, the layout, the `npm run
generate`/`check`/`validate`/`format` workflow, the
`schema.ts`→`schema.json` regeneration rule, the `$schema` URL
convention, the graduation plan), and `AGENTS.md` points to it rather
than restating it. `AGENTS.md` adds only the **contribution conduct the
README doesn't cover**:
- **Open PRs for human review — do not self-merge AI-generated
changes**, even with merge permissions.
- Keep changes small and scoped; one coherent change per PR; link the
relevant SEP/issue.
- Respect the experimental, SEP-2127-tracked nature; don't pre-empt
unresolved spec discussion; keep the graduation path intact.
- Treat the schema as a public contract; flag breaking changes
explicitly.
`AGENTS.md` is the portable, tool-agnostic entry point; `CLAUDE.md` is a
symlink to it so Claude Code loads it automatically — one file, two
entry points, zero duplication.
## Verification
- [x] **`npm run check` passes** (schema.json in sync, `tsc --noEmit`
clean): `✓ schema.json is up to date`
- [x] **`npm run validate` passes**: `All 7 example(s) passed.`
- [x] **`npm run format:check` passes** (the new `AGENTS.md` is
Prettier-clean): `All matched files use Prettier code style!`
- [x] **Change is docs-only** — adds `AGENTS.md` + `CLAUDE.md` symlink;
no `schema.ts`/`schema.json`/examples touched
- [x] **`CLAUDE.md` is a symlink to `AGENTS.md`** (verified `ls -l`:
`CLAUDE.md -> AGENTS.md`), no content duplication
- [x] **No README duplication** — `AGENTS.md` references `README.md` for
all mechanics and only adds contribution conduct (verified by re-reading
both files side by side after a fresh-eyes review)
- [x] **CI green on this PR**: `build`, `CodeQL`, and `Analyze
(actions)` all pass (re-confirmed after each revision)
---
🤖 Opened by an AI coding agent for human review — **not** to be
self-merged (per the contribution norms in this very PR). Please review
and merge if it looks good.
This commit is contained in:
parent
ad000df044
commit
7c340f43e7
2 changed files with 32 additions and 0 deletions
31
AGENTS.md
Normal file
31
AGENTS.md
Normal file
|
|
@ -0,0 +1,31 @@
|
|||
# AGENTS.md — MCP Server Cards (Experimental Extension)
|
||||
|
||||
Operating manual for AI coding agents working in this repository.
|
||||
|
||||
**Read [`README.md`](./README.md) first.** It is the source of truth for everything mechanical — what a Server Card is, the repo layout, the schema build/validate workflow (`npm run generate` / `check` / `validate` / `format`), the rule that `schema.ts` is authoritative and `schema.json` is generated (never hand-edited, regenerated in the same commit, enforced by CI), the `$schema` URL convention, and the graduation plan. This file does **not** repeat any of that; it only adds the contribution conduct the README doesn't cover.
|
||||
|
||||
## Contribution principles
|
||||
|
||||
Ordered most-important first.
|
||||
|
||||
### 1. Open PRs for human review — do not self-merge agent-generated changes
|
||||
|
||||
Agent-driven changes should be opened as a pull request for a human maintainer to review and merge. Even when the account running the agent has merge permissions, **do not merge your own AI-generated PRs** (and do not enable auto-merge). The decision to merge belongs to a human.
|
||||
|
||||
### 2. Stay small, scoped, and conservative
|
||||
|
||||
This is an active, spec-tracked, community project. Make one coherent change per PR; don't bundle unrelated refactors or repo-wide reformatting into a feature/fix. Write clear PR descriptions that link the relevant SEP/issue and explain how you verified the change. Match existing conventions and defer to maintainers on direction. For anything larger than a self-contained change, prefer opening an issue or a draft PR for discussion over a large unilateral edit.
|
||||
|
||||
### 3. Respect the experimental, spec-tracked nature
|
||||
|
||||
Schema decisions here feed SEP-2127 and, eventually, the core MCP spec. Don't make changes that pre-empt or contradict unresolved spec discussion, and keep the README's "Graduation plan" migration path intact. "Experimental" means provisional, not low-stakes.
|
||||
|
||||
### 4. Treat the schema as a public contract
|
||||
|
||||
`schema.json` is consumed by external tools and published documents. If a change is breaking for consumers (removing/renaming fields, tightening required-ness), call it out explicitly in the PR description rather than shipping it silently.
|
||||
|
||||
## What NOT to do
|
||||
|
||||
- Do **not** merge your own AI-generated PR, even with permission (Principle 1).
|
||||
- Do **not** bundle unrelated changes, reformat the whole repo, or do sweeping renames in a feature/fix PR.
|
||||
- Do **not** make schema changes that pre-empt unresolved SEP-2127 / core-spec discussion.
|
||||
1
CLAUDE.md
Symbolic link
1
CLAUDE.md
Symbolic link
|
|
@ -0,0 +1 @@
|
|||
AGENTS.md
|
||||
Loading…
Reference in a new issue