0.2.1 restored Codex boot by flipping required=false. Once @agentsid/
guard@0.1.1 actually landed on npm, the MCP reachability was fixed —
but a new failure surfaced: "MCP client for agentsid timed out after
10 seconds."

Root cause: `startup_timeout_sec = 10` was Codex's own default, which is
too tight for a cold `npx -y @agentsid/guard`. The npx spawn has to
fetch the tarball + @modelcontextprotocol/sdk (~2MB combined) before the
guard process can print its "running" line. On a fresh npx cache that
takes 15-25s; on a warm cache it's <1s.

Live-confirmed: after warming the npx cache manually with a 20s run,
`npx -y @agentsid/guard < /dev/null` emits "AgentsID Guard running"
within a fraction of a second. The guard itself is fine; the timeout
window was just too small for first-run installs.

Bump to 30s — enough headroom for a cold fetch on typical connections
while still fast enough that a genuinely stuck server shows up as a
failure within a reasonable wait. Power users with slow networks can
override in `~/.codex/config.toml`.

Regression-locked via a test that asserts startup_timeout_sec >= 30.