chore: maintain security-sensitive UI surfaces

[myth05]

Summary

• Refresh security-sensitive JS dependency pins/locks across the desktop, TUI, website, and web workspaces, including React Router and DOMPurify remediation coverage.
• Add launchd file descriptor limits for the macOS gateway service while preserving upstream unconditional KeepAlive behavior.
• Add a desktop Company Brain trace parser/formatter primitive for future important-answer source/correction gating.
• Clean up small TypeScript compatibility issues in desktop/TUI/website surfaces.

Verification

• python -m pytest -o 'addopts=' tests/hermes_cli/test_gateway_service.py::TestLaunchdServiceRecovery::test_launchd_plist_sets_file_descriptor_limits tests/hermes_cli/test_gateway_service.py::TestServiceWorkingDirIsStable::test_launchd_plist_keepalive_unconditional -q
• (cd apps/desktop && npx vitest run --environment jsdom src/lib/company-brain-trace.test.ts src/components/pane-shell/pane-shell.test.tsx)
• npm --prefix ui-tui test -- --run packages/hermes-ink/src/utils/execFileNoThrow.test.ts
• npm --prefix apps/desktop run type-check
• npm --prefix ui-tui run type-check
• npm --prefix website run typecheck
• npm audit --audit-level=moderate --json → 0 vulnerabilities

Notes

• Full npm --prefix apps/desktop run test:ui -- company-brain-trace.test.ts currently runs the whole desktop src suite because the script includes src; after merging latest main, that broader suite still has unrelated upstream UI test expectation failures. The targeted new/changed desktop tests pass via direct Vitest invocation above.

[myth05]

CI is currently fork-gated (action_required across workflows), not failing from executed logs.

Local verification completed on the pushed head 5401ca6b84ad6965ef06af014629ecf5bcc66379:

• python -m pytest -o 'addopts=' tests/hermes_cli/test_gateway_service.py::TestLaunchdServiceRecovery::test_launchd_plist_sets_file_descriptor_limits tests/hermes_cli/test_gateway_service.py::TestServiceWorkingDirIsStable::test_launchd_plist_keepalive_unconditional -q → 2 passed
• (cd apps/desktop && npx vitest run --environment jsdom src/lib/company-brain-trace.test.ts src/components/pane-shell/pane-shell.test.tsx) → 19 passed
• npm --prefix ui-tui test -- --run packages/hermes-ink/src/utils/execFileNoThrow.test.ts → 4 passed, 1 skipped
• npm --prefix apps/desktop run type-check → passed
• npm --prefix ui-tui run type-check → passed
• npm --prefix website run typecheck → passed
• npm audit --audit-level=moderate --json → 0 vulnerabilities

Maintainers: please approve the fork workflows when convenient. Maintainer edits are enabled.

[austinpickett]

This PR bundles four unrelated workstreams under a single "chore: maintain security-sensitive UI surfaces" label, and one of them is speculative dead code. It needs to be split before any part can merge.

What's actually in here (14 files, +754/-327)

1. Dependency pin/lock refreshes (root package.json/package-lock.json + website + web + ui-tui + desktop) — legitimate security intent (React Router, DOMPurify), but see conflicts below.
2. launchd file-descriptor limits in hermes_cli/gateway.py — plausibly useful, but unrelated to "UI surfaces."
3. 🔴 apps/desktop/src/lib/company-brain-trace.ts — a new trace parser/formatter primitive. It is imported nowhere (I checked: only its own test references it), and the PR description states it's "for future important-answer source/correction gating." This is speculative infrastructure with no current consumer — we don't merge hooks/primitives until something actually uses them.
4. pane-shell.tsx — a resize-behavior change (drops the pane.resizable guard on widthOverride), unrelated to security; plus misc TS cleanups in tui/website.

Blocking issues

• Speculative code (#3) should be dropped from this PR entirely. If the Company Brain gating feature is coming, land the primitive with its consumer in a dedicated feature PR.
• Conflicts: mergeable: CONFLICTING/DIRTY. The root package.json/package-lock.json edits also overlap several in-flight lockfile PRs — #44024 (@types/node, approved), #44197 (icons-pack, approved), and #39450 (website/whatsapp-bridge audit). These can't all merge independently; the dep work here should be reconciled with those rather than duplicating them.
• Misleading label: a "chore" touching gateway.py service config, adding a new desktop primitive, and changing pane resize behavior isn't a chore.

Ask

Please split into focused PRs, each rebased on current main:

1. The dependency/lock refresh — but coordinate with #44024 / #44197 / #39450 so it's only the net-new remediation, not a re-do of what's already approved.
2. The launchd FD-limit change (its own fix(gateway) PR with the test you already wrote).
3. Drop company-brain-trace.ts until it has a real consumer.
   The pane-shell tweak, if intentional, belongs in whichever UI PR actually needs it. Happy to fast-track the launchd and dep pieces once separated.

[myth05]

Thanks, Austin. Split out the launchd piece into a focused PR from current main:

• fix(gateway): set launchd file descriptor limits #44357 — fix(gateway): set launchd file descriptor limits

It contains only hermes_cli/gateway.py plus the existing focused test. Verified locally:

uv sync --extra dev
uv run --extra dev python -m pytest -o 'addopts=' tests/hermes_cli/test_gateway_service.py::TestLaunchdServiceRecovery::test_launchd_plist_sets_file_descriptor_limits tests/hermes_cli/test_gateway_service.py::TestServiceWorkingDirIsStable::test_launchd_plist_keepalive_unconditional -q\n# 2 passed\n```\n\nI am not carrying over the Company Brain primitive or pane-shell tweak. I am leaving the dependency work out of this split until it can be reconciled against #44024 / #44197 / #39450 instead of redoing their lockfile changes.

[teknium1]

Thanks for the security-maintenance work. The remaining useful pieces need to stay split rather than landing as this bundle.

Problems

• apps/desktop/src/lib/company-brain-trace.ts has no production consumer in this PR: its only added import is its own test, while the body describes future gating. AGENTS.md:98-101 rejects speculative infrastructure without a concrete consumer.
• The execFileNoThrow typing change is already on current main in 4ce9caed0 (ui-tui/packages/hermes-ink/src/utils/execFileNoThrow.ts:1,35,37), and React Router was already remediated in 46c16b928 (apps/desktop/package.json:103, web/package.json:34).
• The dependency work is not wholly obsolete: current website/package-lock.json:9311 still resolves DOMPurify 3.4.2, so any website remediation must be regenerated against current lockfiles.

Suggested changes

• Keep the launchd limits in focused PR #44357.
• Drop the unconsumed Company Brain parser and its tests.
• Re-scope remaining dependency work to unresolved current-main packages; retain the pane-shell change only as a separately justified desktop fix.

Automated hermes-sweeper review.

[teknium1]

Please remove this parser from this PR. The diff adds no production import or consumer—only its own test—and the stated use is future gating; AGENTS.md:98-101 rejects speculative infrastructure without a concrete consumer.

[teknium1]

This exact ChildProcess/StdioOptions typing change is already present on current main in commit 4ce9caed0 (fix(tui): type execFileNoThrow stdio/ChildProcess). Omit it from any salvage.

[myth05]

Closing this broad PR as superseded after maintainer review. The useful launchd file-descriptor change was split into focused PR #44357. I am dropping the unconsumed Company Brain parser and pane-shell change, and I will not carry forward the stale/overlapping dependency edits. Any future dependency remediation will be regenerated as a separate focused PR from current main.