test(server): pin the console collector contract - ConsoleCollector (Closes #1663)

[dmitrii-f-t27]

Closes #1663

What changed

One new file: trios/agent-server/apps/server/tests/api/console-collector.test.ts (the path the issue's Success Criteria names). The subject browser/console-collector.ts is untouched.

The single exported class is exercised (10 it( blocks) through a recording fake CdpBackend whose handlers the suite invokes directly:

• Event routing (5 tests): console types mapped to levels with args serialized space-joined (measured: a trailing-space fixture would double the join); unknown types fall back to info; exceptions recorded as errors with the description preferred over the text; browser log entries by the log level map (verbose -> debug); a main-frame navigation clears the buffer while a child frame does not.
• Sessions (3 tests): events from unknown sessions are ignored; re-attach replaces the old session; detach forgets the page.
• getLogs (2 tests): the severity threshold (measured: the default is 'info', so debug-level entries are hidden unless asked for), the case-insensitive search, combined filters; and the recency limit (60 entries -> the last 50, the honest totalCount of 60) with the optional clear.

The suite registers no mocks. No network, database or container.

How it was verified (FR-004)

• Passing run against the unmodified subject: 10 pass / 0 fail / 26 expect() calls, Ran 10 tests across 1 file.
• Deliberately broken copy (the console type map remapped, args joined with '-', the severity comparison inverted, re-attach stops replacing the old session, child frames clear too): 6 fail out of 10, quoted raw above. The subject was restored byte-identical (git checkout) and the suite passes again.
• All shapes were measured on the live module before writing (two drafts were corrected by measurement: the default 'info' threshold and the double-space join). The file is biome ci-clean (checked with an explicit Found/error grep - the noNonNullAssertion warnings of the first draft led to the fire() helper) and contributes no typecheck error (local monorepo typecheck: base set, 0 mentions of the new file). The one-process run with the four sibling api suites is 41/41; the full tests/api run shows only the 8 pre-existing local vault failures (age is not installed on this machine - CI installs it, see PR ci(test): install age where the vault's tests start it, before the vault ship #1782).

Not verified

• Nothing beyond the quoted local runs; the CI Tests / server-api job re-runs the suite on this SHA.

Dmitrii Fedorov (@dmitrii-f-t27)

[github-actions]

✅ Tests passed — 3482/3552

Suite	Passed	Failed	Skipped
✅ agent	144/144	0	0
✅ build	9/9	0	0
✅ cdp-protocol	5/5	0	0
✅ eval	93/93	0	0
✅ server-agent	280/280	0	0
✅ server-api	2125/2194	0	69
✅ server-browser	6/6	0	0
✅ server-integration	10/11	0	1
✅ server-lib	279/279	0	0
✅ server-pglive	166/166	0	0
✅ server-root	68/68	0	0
✅ server-sim	8/8	0	0
✅ server-skills	31/31	0	0
✅ server-tools	244/244	0	0
✅ shared	14/14	0	0

View workflow run

[dmitrii-f-t27]

Checks note, before merge:

• runner / Biome and runner / Typecheck are red on the base branch queen with pre-existing failures (reference base run https://github.com/gHashTag/trios/actions/runs/38087839871 on dfb6d97). On this PR's SHA: the Typecheck log has the 19 base error TS lines and 0 mentions of the new console-collector.test.ts; the Biome log has 0 diagnostics for the new file (the first draft's noNonNullAssertion findings were fixed before pushing - the suite invokes the fake's handlers through a fire() helper). The suite itself passed in Tests / server-api on this SHA (10 tests), and all other Tests / * suites including server-pglive are green.

Dmitrii Fedorov (@dmitrii-f-t27)