fix: harden gateway status and Photon sidecar startup

[nomadjames]

Summary

• Preserve the Claude handoff memory endpoint on the API server.
• Add safer gateway long-turn notices and turn-usage reporting.
• Preserve explicit zero budget settings in gateway usage handling.
• Tighten profile gateway/status detection, including systemd profile services, bare-list cron stores, and host-vs-container detection.
• Keep the Photon sidecar alive if the optional Spectrum mixed-attachment compatibility patch fails, so basic iMessage transport is not taken down by patch/package-layout drift.

Test plan

• node --check plugins/platforms/photon/sidecar/index.mjs
• node --check plugins/platforms/photon/sidecar/patch-spectrum-mixed-attachments.mjs
• ./venv/bin/python -m py_compile gateway/platforms/api_server.py gateway/run.py gateway/long_run_status.py gateway/turn_usage.py gateway/slash_commands.py hermes_cli/config.py hermes_cli/gateway.py hermes_cli/status.py hermes_constants.py plugins/platforms/photon/adapter.py
• ./venv/bin/python -m pytest -q tests/gateway/test_api_server.py tests/gateway/test_api_server_claude_handoff.py tests/gateway/test_gateway_long_running_notifications.py tests/gateway/test_long_run_status_enrichment.py tests/gateway/test_turn_usage.py tests/gateway/test_usage_command.py tests/hermes_cli/test_gateway.py tests/hermes_cli/test_gateway_service.py tests/hermes_cli/test_status_redaction.py tests/plugins/platforms/photon/test_mention_gating.py tests/plugins/platforms/photon/test_spectrum_patch.py tests/test_hermes_constants.py -o 'addopts='

Local result: 513 passed, 112 warnings.

[tonydwb]

Code Review Summary

Observation: High surface area (22 files, +1711/-44)

This PR introduces a new 365-line module (gateway/long_run_status.py) for optional gateway status enrichment, plus changes to gateway/run.py for noisy status filtering and a new handle_busy_turn_choice method.

Positive observations:

• The enrichment module is deliberately fail-soft (docstring is clear)
• Status line validation regex looks correct
• The _GATEWAY_NOISY_STATUS_RE expansion to cover more platforms is a good improvement
• The handle_busy_turn_choice method is well-structured with clear branching

Areas for human review:

• 22 files is above the surface-area threshold for automated confidence
• The new long_run_status.py module integrates with external services (Ollama, OpenRouter) — worth verifying the error handling covers all failure modes
• The busy-turn choice feature touches gateway run.py core — confirm no prompt-caching impact

Recommend human review before merge due to surface area.

[teknium1]

Thanks for addressing several real gateway and Photon failure modes. The Photon premise is confirmed on current main: plugins/platforms/photon/sidecar/index.mjs:225-233 still exits the sidecar when the optional compatibility patch fails.

Problems

• gateway/platforms/api_server.py:1292-1297 advertises a handoff API but imports clarence_db by injecting ~/.openclaw/workspace/scripts into global sys.path. The repository has no clarence_db implementation or declared dependency, and tests/gateway/test_api_server_claude_handoff.py:61-74 succeeds only by injecting a fake module. A standard installation would return the endpoint's 500 path.
• gateway/long_run_status.py:318 records the daily OpenRouter call before urlopen() succeeds (:320-323), so transient failures consume the quota.

Suggested changes

• Put the handoff write behind a declared, configured Hermes memory-provider surface and advertise/register it only when available; avoid request-time global sys.path mutation.
• Record cloud usage only after a successful response, with a failed-request quota regression test.

Current main also still waits 180 seconds before the first long-turn notice (gateway/run.py:19336-19342), so the core notification and Photon fixes remain worth salvaging. This is an automated hermes-sweeper review.

[teknium1]

This advertised endpoint depends on an undeclared deployment-local module loaded by mutating global sys.path; the repository contains no clarence_db package or dependency, while the test injects a fake module. Please use a declared/configured Hermes memory-provider integration and gate the capability/route when it is unavailable.

[teknium1]

This increments the daily quota before urlopen() succeeds, so timeouts and transport failures consume the allowance. Record the call only after a successful response is accepted, and add a failed-request regression test.