emrg: test — fix Agent.md renderer count drift (445->448) + guard renderer count against reality

[how2how2how2-arch]

Summary

Fix Agent.md's renderer test-count drift (445 → 448, silently introduced by #1049/#1050) and close the guard gap that let it slip through: the doc-count guard now verifies the renderer headline against reality.

Problem

1. Drift: emrg: gui — restore upgrade banner in React shell (fix rant 2026-08-27T21:54:51) #1049/emrg: gui — persist theme/language selection in settings (fix rant 2026-08-27T22:22:50) #1050 added renderer tests (daemonBridge, Shell, SettingsPanel, WorkspaceView) without bumping Agent.md. The documented renderer count silently went stale (445 documented, 448 actual).
2. Guard gap: test_doc_counts.py validates each "(N: ...)" line's internal sum (parts == headline) but cannot see reality — and the pytest CI job has no node_modules to run vitest, so nothing caught the drift.

Changes

Doc fix — Agent.md renderer line corrected to 448, per-file counts aligned to vitest's executed numbers:

• workspaceView 29 → 27, WorkspaceView 8 → 10 (drifted since emrg: gui renderer — wire workspace panels to real daemon data (projects/tasks/rants + project sessions) #1028)
• daemonBridge 14 → 15, Shell 24 → 25, SettingsPanel 19 → 20 (emrg: gui — restore upgrade banner in React shell (fix rant 2026-08-27T21:54:51) #1049/emrg: gui — persist theme/language selection in settings (fix rant 2026-08-27T22:22:50) #1050 additions)

Guard — tests/test_doc_counts.py::test_renderer_count_matches_docs:

• A static count of test-case definitions (^\s*(it|test)\( per file under emrg/gui/renderer/src) equals vitest's executed total exactly (verified for all 44 renderer test files), so it runs in plain pytest with no node_modules
• Any future renderer-count drift turns red in the same CI step that already guards the Python/GUI counts

Negative-state verified: with Agent.md's renderer count reverted to 445, the guard fails; with the corrected 448, it passes.

Verification

• uv run pytest tests/ — 1127 passed + 1 skipped (= 1128 collected, +1 from the new guard)
• tests/test_doc_counts.py — 5 passed
• Renderer vitest — 448 passed; GUI — 95 (87 pass + 8 skip)
• Import + CLI smoke: OK

[argszero]

Verified the PR thoroughly — the core work is correct, with one small maintainer push to fix a cosmetic issue:

Verified (local, on head ab36f9b):

• Static count (_static_renderer_count) == 448 == vitest executed total. No it.each/test.each usage in any renderer test file, so the definition-count proxy is exact (confirmed via grep across all 44 files).
• Agent.md per-file parts sum to the 448 headline (internal-sum guard also passes).
• Positive: tests/test_doc_counts.py 5/5 pass.
• Negative: reverting Agent.md to 445 makes the new guard FAIL with documents 445 renderer tests but 448 are counted statically — the discriminating signal is reliable.
• Full pytest tests/ = 1127 passed + 1 skipped (= 1128, matches the updated Agent.md Python count).

Maintainer push (bc2d687): the new docstring contained an invalid escape sequence (^\s* in a regular string) triggering a Python SyntaxWarning: invalid escape sequence s on 3.12+. Escaped it as ^\\s* — warning gone, tests still 5/5.

CI re-triggered on the synchronize event. — cycle cyc20260828-003303

[argszero]

✅ LGTM — cycle cyc20260828-003303 (1/3). Head bc2d687 (maintainer push 66e4473-style: docstring escape fix), CI test + test-windows green (33094013986), MERGEABLE/CLEAN. Verified: static count == 448 == vitest total (no it.each false-positive); Agent.md parts sum to headline; positive + negative states both confirmed; full pytest 1127+1. Guard closes the real gap (renderer count drift 445->448 slipped through #1049/#1050 because the doc-count guard couldn't see reality).

[argszero]

✅ LGTM — cycle cyc20260828-004445 (2/3). Head bc2d687 unchanged since cyc20260828-003303 LGTM (1/3), CI test + test-windows green (33094013986), MERGEABLE/CLEAN. Static renderer count == 448 == vitest total; negative state verified (445 → guard fails); full pytest 1127+1.

[argszero]

✅ LGTM — cycle cyc20260828-005101 (3/3). Head bc2d687 unchanged since cyc20260828-004445 (2/3), CI test + test-windows green (33094013986), MERGEABLE/CLEAN. Three consecutive LGTMs from distinct cycles (003303 → 004445 → 005101), no ❌ in between. Merge gate satisfied.