docs(skills): clarify Bun audit and CI timing guidance

.agents/skills/bun-runtime/SKILL.md

@@ -24,11 +24,11 @@ Use when: adopting Bun, migrating from Node, writing or debugging Bun scripts/te
 
 **Migration from Node**: Replace `node script.js` with `bun run script.js` or `bun script.js`. Run `bun install` in place of `npm install`; most packages work. Use `bun run` for npm scripts; `bun x` for npx-style one-off runs. Node built-ins are supported; prefer Bun APIs where they exist for better performance.
 
-**Package and workspace commands**: `bun add <pkg>` / `bun remove <pkg>` / `bun update [pkg]` manage dependencies; `bun outdated` lists stale deps; `bun pm ls` shows the dependency tree derived from the lockfile (not a listing of what's actually on disk). Read-only inspection: `bun audit` reports known vulnerabilities. Modifying: `bun audit fix` applies fixes, `bun dedupe` collapses duplicate installs, and `bun prune` removes packages no longer referenced. Monorepos use a root `workspaces` array in `package.json` (same convention as npm/yarn); run a script in one workspace with `bun run --filter <pkg-name> <script>`.
+**Package and workspace commands**: `bun add <pkg>` / `bun remove <pkg>` / `bun update [pkg]` manage dependencies; `bun outdated` lists stale deps; `bun pm ls` shows the dependency tree derived from the lockfile (not a listing of what's actually on disk). Read-only inspection: `bun audit` reads `bun.lock` and sends the package list to the npm advisory endpoint (or each scoped registry's own advisory endpoint) to report known vulnerabilities — it doesn't touch local files, but it isn't offline, so on private projects only use an approved registry/advisory endpoint, since package names and versions (including private ones) leave the machine. Modifying: `bun audit fix` applies fixes, `bun dedupe` collapses duplicate installs, and `bun prune` removes packages no longer referenced. Monorepos use a root `workspaces` array in `package.json` (same convention as npm/yarn); run a script in one workspace with `bun run --filter <pkg-name> <script>`.
 
 **Built-in APIs**: reach for these before adding a dependency — `Bun.file` / `Bun.write` for file I/O, `bun:sqlite` for an embedded SQLite database, `Bun.serve` for an HTTP/WebSocket server, [`Bun.sql`](https://bun.com/docs/runtime/sql) for SQL databases, [`Bun.redis`](https://bun.com/docs/runtime/redis) for Redis, and [`Bun.S3Client`](https://bun.com/docs/runtime/s3) for S3-compatible object storage. See the [1.4 release notes](https://bun.com/blog/bun-v1.4) for the rest of what's new.
 
-**Vercel**: Set runtime to Bun in project settings. Build: `bun run build` or `bun build ./src/index.ts --outdir=dist`. Install: `bun install --frozen-lockfile` for reproducible deploys.
+**Vercel**: Set `bunVersion: "1.4.x"` in `vercel.json` to use Bun 1.4 (Rust runtime) — `"1.x"` currently still resolves to the older 1.3.14. See [Vercel's Bun runtime docs](https://vercel.com/docs/functions/runtimes/bun). Build: `bun run build` or `bun build ./src/index.ts --outdir=dist`. Install: `bun install --frozen-lockfile` for reproducible deploys.
 
 **Reference**: [Bun 1.4 release notes](https://bun.com/blog/bun-v1.4).
 
@@ -60,7 +60,7 @@ bun test
 bun test --watch
 ```
 
-For CI, Bun 1.4's test runner adds: `bun test --changed[=<ref>]` (run test files affected by changes vs. a git commit/branch — includes tests exercising changed source files, not just edited test files), `bun test --isolate` (fresh global object per test file, so leaked handles in one file can't affect another; implied by `--parallel`), `bun test --parallel[=<n>]` (run test files across N worker processes, defaults to CPU core count), `bun test --shard=<n>/<count>` (split one suite across CI runners), and `bun test --timings=<file>` with `--update-timings` (merge per-file duration data across shards so `--shard`/`--parallel` can front-load the slowest files). Example: `bun test --parallel --timings=./test-timings.json --update-timings`. Run `bun test --help` locally to confirm exact flag behavior for your installed version.
+For CI, Bun 1.4's test runner adds: `bun test --changed[=<ref>]` (run test files affected by changes vs. a git commit/branch — includes tests exercising changed source files, not just edited test files), `bun test --isolate` (fresh global object per test file, so leaked handles in one file can't affect another; implied by `--parallel`), `bun test --parallel[=<n>]` (run test files across N worker processes, defaults to CPU core count), `bun test --shard=<n>/<count>` (split one suite across CI runners), and `bun test --timings=<file>` (reads JSON per-file duration data to front-load the slowest files) with `--update-timings` (writes to the first `--timings` path given — without `--shard` it merges into existing entries; with `--shard` it writes only the current shard's files to that same first `--timings` path — Bun does not pick a different path per shard automatically, so choose a distinct output path for each shard to avoid overwriting other shards' timing records; pass `--timings` multiple times, once per shard file, to read them together as a single combined table). Example: `bun test --parallel --timings=./test-timings.json --update-timings`. Run `bun test --help` locally to confirm exact flag behavior for your installed version.
 
 ```typescript
 // test/example.test.ts