feedback7: AGENTS.md and CLAUDE.md — an entry point for AI coding assistants (GitHub #379 by @anAtheist987; the owner's OK of 2026-10-08)

- AGENTS.md only indexes the existing documents (ARCHITECTURE, DESIGN index, SIM / META / DATA / I18N, the kits README)
  and their hard rules (official first + [ASSUMED], sim determinism and detmath, golden results, generated data, no game
  art, t() strings, docs with the code, commit / PR conventions, verification commands); CONTRIBUTING.md stays the full
  text. CLAUDE.md is one line pointing to it.
- Tests: docs-paths, docs-consistency.

AGENTS.md

@@ -0,0 +1,56 @@
+# AGENTS.md
+
+Entry point for AI coding assistants (Codex, Claude Code, Cursor, Copilot …) working on this repository. It only
+indexes the existing documents and their hard rules; when this file and a linked document disagree, the document wins.
+Human contributors: [CONTRIBUTING.md](CONTRIBUTING.md) is the same material in full.
+
+## What this is
+
+A non-commercial fan remake of Arknights「卫戍协议：盟约」 that runs in the browser: a Node.js server (economy, rounds,
+rooms) and a deterministic battle sim shared by the server and the browser. It aims to be faithful to the official mode.
+
+## Read first
+
+1. [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md) — the code map: where things live, data flow, golden results,
+   import boundaries, where common changes start.
+2. [docs/DESIGN.md](docs/DESIGN.md) — the index of the rules: current rules in `docs/design/`, each release's
+   revisions and their evidence in `docs/history/`.
+3. The reference for the part you touch: [docs/SIM.md](docs/SIM.md) (battle engine API, hooks, test harness),
+   [docs/META.md](docs/META.md) (match flow, shop, protocol), [docs/DATA.md](docs/DATA.md) (generated data),
+   [docs/I18N.md](docs/I18N.md) (interface strings), `server/sim/content/kits/README.md` (operator kits).
+
+## Hard rules
+
+- **Official first.** Rules come from the official data tables (`.cache/gamedata/excel/` after setup) and
+  [PRTS](https://prts.wiki/). A detail no source settles is implemented the simplest way and marked `[ASSUMED]` in the
+  code and the PR. Deliberate deviations from the official mode are the maintainer's decision only.
+- **Determinism.** `server/sim/` is plain ESM that runs in Node and in browsers: no Node APIs, no `Math.random`, no
+  clocks, and no engine-approximated Math functions (use `server/sim/detmath.js`; ESLint and
+  `test/sim/detmath.test.js` enforce it).
+- **Golden results.** A refactor must not change `test/golden/*.json`. An intended gameplay change runs
+  `npm run golden:update` in the same commit, and the commit / PR names every moved scenario and why
+  ([test/golden/README.md](test/golden/README.md)).
+- **Generated data.** Never hand-edit `data/*.json`: change `tools/build-data.mjs` and regenerate.
+- **No game art in git.** `public/assets/` is ignored; never commit extracted or downloaded game files.
+- **Interface strings** go through `t('…')`, with entries in every pack under `public/i18n/`
+  (`node tools/i18n.mjs check --all --strict`).
+- **Docs follow the code.** A rule change updates `docs/design/`, the current `docs/history/` file and the reference
+  docs; run `node --test test/docs-consistency.test.js test/docs-paths.test.js`.
+- **Commits and PRs**: one topic per PR; a one-line message of what changed (Chinese or English) with the issue number;
+  no attribution trailers (`Co-Authored-By:`, `Generated with …`). The PR description states what, why, the sources,
+  the `[ASSUMED]` points, the tests you ran (paste the summary lines) and the golden moves. No ads, payments or other
+  monetisation — the project stays non-commercial.
+
+## Verify
+
+Run the tests of what you touched first, then everything before opening the PR:
+
+```bash
+node --test test/<area>/<file>.test.js      # targeted
+npm run ci                                  # the CI checks locally (tools/ci.mjs): node --test, smoke, lint, imports, types
+npm run golden                              # golden results
+```
+
+Browser suites are off by default and need Chrome: `SP_E2E=1` (UI), `SP_REAL_E2E=1` (real match, needs assets),
+`RENDER_E2E=1`, `SIM_E2E=1` — see CONTRIBUTING.md §2. Battle tests are easiest with the helpers described in
+docs/SIM.md (the test harness section) and the patterns in `test/content/op_siege.test.js`.

CLAUDE.md

@@ -0,0 +1,3 @@
+# CLAUDE.md
+
+Read [AGENTS.md](AGENTS.md) — the entry point for AI coding assistants (what to read first, the hard rules, how to verify).