Sync upstream/main (96 commits)

[randlee]

Automated fork sync: merge upstream/main (NousResearch/hermes-agent) into randlee/hermes-agent.

Changes: 212 files changed, +16,045 / -2,929

Key changes from upstream:

• gateway/run.py: multiplex busy input mode, relay followup/teardown contracts
• hermes_cli/: kanban review lifecycle, dashboard, resource limits, plugins system
• tools/: browser_use_cli, kanban tools improvements
• skills/: new merge-reconciler and sdlc-review skills
• apps/desktop/: window renderer lifecycle, renderer logging
• pyproject.toml: python cap raised to <3.15

Generated by: grecon fork-sync pipeline

[github-actions]

૮ >ﻌ< ა ci review

_{running on 5aa73db — Sync upstream/main (96 commits)}

_{waiting for jobs to start…}

[randlee]

Contessa Review — Fork Sync PR #2

CI Results

Check	Status
Python tests (11/12 slices)	✅ Pass
Python tests (slice 3/12)	❌ FAIL
JS/TS checks (all)	✅ Pass
Python lints	✅ Pass
Docs checks	✅ Pass
Installer tests	✅ Pass
Supply-chain scan	✅ Pass

Failing Test

tests/gateway/test_multiplex_busy_input_mode.py::test_profile_route_and_nonmultiplexed_resolution_preserve_boundaries

AssertionError: assert 'interrupt' == 'steer'

Root cause: Rejecting profile route 'research-chat': target profile 'research' is not served

Analysis

• This is an upstream bug, not fork-introduced. The test file was created in upstream commit a31be4803 ("fix(gateway): respect routed profile busy modes"). The same test fails on NousResearch/hermes-agent/main.
• 14 of 15 tests in that file pass; 1 assertion is wrong in a profile routing boundary check.
• The merge itself is clean: 212 files, no conflicts, no manual resolution needed.

Verdict: NEEDS_FIX — test failure in upstream code

• Does NOT block merge — the fork faithfully mirrors upstream state including this known bug.
• Recommend: merge as-is and track upstream fix. The test failure exists in CI on NousResearch/hermes-agent too.

[randlee]

Alpha-Prime Review — PR #2

Investigation of test failure

Failing test: test_profile_route_and_nonmultiplexed_resolution_preserve_boundaries

Root cause analysis: The test at line 339 expects _effective_busy_input_mode(source) to return "steer" (from the research profile snapshot), but it returns "interrupt" (the default). The call chain:

1. _effective_busy_input_mode → _busy_profile_name_for_source → _profile_name_for_source
2. _profile_name_for_source matches the route for research-chat → profile "research"
3. It then checks if "research" is in the served-profiles set (line 25429: _multiplex_profile_homes(config))
4. Since "research" is NOT served → ProfileRouteRejected → caught → returns None
5. Fallback: "interrupt" is used

The test needs runner.config.multiplex_profile_homes (or whatever _multiplex_profile_homes reads) to include "research". The test author only called _snapshot_profile_busy_modes but didn't configure the profile as served.

Decision: NO FIX IN FORK

This is an upstream bug in commit a31be4803. The same test fails on NousResearch/hermes-agent/main. Fork-sync must faithfully mirror upstream — fixing upstream bugs here creates merge divergence.

Recommendation: Merge as-is. The fork correctly reflects upstream state. The test failure is cosmetic (1 of 15 tests in a new file, does not affect functionality). File upstream issue if desired.

[randlee]

Superseded: the fork now syncs via REBASE (main = upstream/main + 1 fork patch) instead of merge-based sync. This merge PR was a stale balloon. Closing.