chore(memory): note the CodeRabbit-vs-tsgolint conflict on Bun rejects assertions

Signed-off-by: Rhuan Barreto <rhuan@barreto.work>

Author: rhuanbarreto

.claude/agent-memory/archgate-developer/MEMORY.md

@@ -33,6 +33,7 @@ Exceptions: minor follow-up tweaks after validation already passed, and non-code
 - **`docs/public/llms-full.txt` is auto-regenerated by the `update-llms.yaml` PR workflow whenever `docs/src/content/docs/**` changes** — a bot commit lands on the branch shortly after pushing docs edits. Never hand-edit this file; `git pull` before continuing work on the branch, and don't be alarmed when CodeRabbit flags stale wording inside it — fix the source `.mdx`, not the generated file.
 - **Sizing a flag-removal's blast radius by grep: search each token separately, don't require them on one line.** A pattern requiring `"check"` and `"--json"` on the same line missed `tests/commands/check.test.ts`, where `c.name() === "check"` and `o.long === "--json"` sit on adjacent lines — the test only surfaced when `bun run validate` failed after the "complete" migration.
 - **`archgate review-context --verbose`'s `decision`/`dosAndDonts` fields are already truncated to the cap plus a fixed-length `[... truncated ...]` marker** — measuring `.length` on them to check whether an ADR edit is under budget returns a constant (cap + marker length) whenever the raw content is still over, masking whether a trim actually helped. Measure the real raw length via `archgate check --output json`'s `briefingWarnings[].length` field instead (only present when over budget; absent once trimmed enough).
+- **CodeRabbit and oxlint-tsgolint give contradictory orders on Bun's `expect(...).rejects.toThrow(...)`** — CodeRabbit flags the unawaited form as a race; the repo's type-aware lint (`await-thenable` + `no-confusing-void-expression`, via bun-types declaring it `void`) hard-rejects the `await`. The lint gate wins: keep the unawaited form (the convention #534 established repo-wide), decline the CodeRabbit finding on-thread citing the lint conflict, and don't re-litigate per PR — fixing it properly is a repo-wide helper/bun-types follow-up.
 - **`review-context` omits every ADR's `decision`/`dosAndDonts` prose by default** — the `@reviewer` skill's own instructions describe the briefing as containing "only the Decision and Do's and Don'ts sections," but the CLI's actual default (confirmed via `--help`) leaves both fields off; only `--verbose` includes them. Without it, per-domain sub-agent prompts built from the response carry `id`/`title`/`domain` and nothing to actually check code against. Always run `bun run cli review-context --verbose` (per-domain, via `--domain <name>`, to keep each call small) when gathering context for the reviewer's sub-agent prompts.
 
 ## Topic files