You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
What happened (live on Jason's main install, 2026-06-11)
Gateway (installed 2026.6.10-dev.1, LaunchAgent ai.argent.gateway) was down all morning. Operator-visible symptom: "I started Argent and it looked like the gateway was down" — Argent.app + dashboard servers up, nothing listening on the configured gateway port (18789), launchd last exit code = 1.
~/.argentos/logs/gateway.log shows the same sequence repeating every boot attempt:
◇ Legacy config keys detected
│ - alwaysOnLoop: alwaysOnLoop was removed; health checks now run directly in the gateway (auto-migrated on load).
◇ Argent repairs
│ Removed deprecated config key: alwaysOnLoop.
◇ Argent doctor
│ Run "argent doctor --fix" to apply legacy migrations.
…while ~/.argentos/argent.jsonstill contained alwaysOnLoop — the boot-time auto-migrate (src/config/legacy.migrations.part-3.tsremove-alwaysOnLoop, ~line 191) mutates the in-memory config and reports "Removed", but the result is never written back to disk, and the process exits 1. Next launch repeats identically. The loop also bloated gateway.log to 755MB.
Manual argent doctor --fix persisted the removal (verified key gone, backup written) and the gateway came up clean immediately after.
Why this is worse than a config nit
The log says the repair happened. "Removed deprecated config key" is a lie at the persistence layer — operator (and any agent reading logs) concludes the key can't be the problem.
Boot-time auto-migrate persists what it migrated (atomic write + .bak, same as doctor --fix), so "auto-migrated on load" is true; OR boot treats legacy keys as non-fatal and proceeds with the in-memory migrated config instead of exiting 1.
If boot still refuses to start for any config reason: surface it to the operator (app banner / argent CLI status / dashboard), not just the log.
Repro
Add "alwaysOnLoop": {} to ~/.argentos/argent.json, start the gateway via LaunchAgent, observe crash loop + log claiming removal.
What happened (live on Jason's main install, 2026-06-11)
Gateway (installed
2026.6.10-dev.1, LaunchAgentai.argent.gateway) was down all morning. Operator-visible symptom: "I started Argent and it looked like the gateway was down" — Argent.app + dashboard servers up, nothing listening on the configured gateway port (18789), launchdlast exit code = 1.~/.argentos/logs/gateway.logshows the same sequence repeating every boot attempt:…while
~/.argentos/argent.jsonstill containedalwaysOnLoop— the boot-time auto-migrate (src/config/legacy.migrations.part-3.tsremove-alwaysOnLoop, ~line 191) mutates the in-memory config and reports "Removed", but the result is never written back to disk, and the process exits 1. Next launch repeats identically. The loop also bloated gateway.log to 755MB.Manual
argent doctor --fixpersisted the removal (verified key gone, backup written) and the gateway came up clean immediately after.Why this is worse than a config nit
doctor --fixlives only in a 755MB log file. This is exactly the Argent runtime stability sweep — subagent dispatch, worker lifecycle, model routing #423 class C gap (runtime degradation surfacing).Wanted (either is fine, both is better)
.bak, same as doctor --fix), so "auto-migrated on load" is true; OR boot treats legacy keys as non-fatal and proceeds with the in-memory migrated config instead of exiting 1.argentCLI status / dashboard), not just the log.Repro
Add
"alwaysOnLoop": {}to~/.argentos/argent.json, start the gateway via LaunchAgent, observe crash loop + log claiming removal.Related: #423 (degradation surfacing class).
🤖 Filed from a live diagnosis with Claude Code