instar 0.28.3 → 0.28.4

package/upgrades/0.28.4.md

@@ -0,0 +1,22 @@
+# Upgrade Guide — v0.28.4
+
+<!-- bump: patch -->
+
+## What Changed
+
+- **Startup grace period for gate evaluation**: Job gates (health checks, quota checks) that require HTTP endpoints are now evaluated after a configurable grace period (default 5 seconds) at startup, rather than immediately. This fixes a race condition where health-gated jobs would permanently fail because the HTTP server wasn't ready yet during the initial missed-job catch-up evaluation. Configurable via startupGraceMs in JobScheduler config. Set to 0 to disable.
+
+- **QuotaTracker stale data handling**: When quota data is older than the staleness threshold (default 30 minutes), the tracker now returns null instead of the stale values. This triggers the existing fail-open path, allowing jobs to run when quota data is unavailable rather than blocking them with outdated percentages.
+
+## What to Tell Your User
+
+- **"Scheduled jobs should start running more reliably"**: If your user noticed that certain scheduled jobs never seemed to execute despite being enabled and due, that should be resolved now. The system was accidentally blocking them during startup.
+
+- **"The system handles missing data more gracefully"**: When usage tracking data gets stale, the system now allows work to continue rather than blocking it based on outdated information.
+
+## Summary of New Capabilities
+
+| Capability | How to Use |
+|-----------|-----------|
+| Startup grace period | Automatic — 5s default delay before first gate evaluation |
+| Stale quota fail-open | Automatic — stale data allows all jobs to proceed |

package/upgrades/0.28.3.md

@@ -1,35 +0,0 @@
-# Upgrade Guide — v0.28.3
-
-<!-- bump: patch -->
-
-## What Changed
-
-- **activationChallenge forwarding**: The `/features/:id/transition` route handler now correctly forwards the `activationChallenge` field when transitioning feature states. Previously, the field was missing from the destructured request body and the `transition()` options, causing activation challenges to silently fail.
-
-- **Slack session continuity**: Fixed an issue where Slack-connected sessions could lose context after compaction or recovery events. Sessions now properly restore their conversation state, preventing the "amnesia" behavior where the agent forgets the ongoing conversation.
-
-- **Supervisor restart cascade prevention**: Added protection against runaway supervisor restart loops under high CPU load. The supervisor now tracks restart frequency and backs off exponentially when it detects a cascade pattern, preventing resource exhaustion.
-
-- **Dashboard session names**: The dashboard now resolves session display names from Telegram topic names instead of showing raw session IDs, making it easier to identify which session is which.
-
-- **Config path respect**: Sessions now properly honor `sessions.claudePath` and `sessions.tmuxPath` from config.json instead of always auto-detecting. This fixes environments where the auto-detected paths were wrong.
-
-- **iMessage mention trigger**: Fixed handling of the `iMessage;-;` prefix in chatId detection so that mention trigger mode no longer incorrectly gates 1:1 conversations. Direct messages now always reach the agent regardless of mention settings.
-
-## What to Tell Your User
-
-- **"Feature activation should work more reliably now"**: If your user was having trouble with feature activation challenges not completing, that should be resolved. The system was accidentally dropping part of the activation handshake.
-
-- **"Slack conversations should keep their context better"**: If your user noticed you forgetting things mid-conversation on Slack, that has been fixed. You will now maintain context even after internal recovery events.
-
-- **"The system is more stable under heavy load"**: Under the hood, the restart behavior is smarter now. Your user does not need to do anything — things should just be more reliable.
-
-## Summary of New Capabilities
-
-| Capability | How to Use |
-|-----------|-----------|
-| activationChallenge fix | Automatic — feature transitions now forward all required fields |
-| Slack session continuity | Automatic — sessions restore context after compaction/recovery |
-| Supervisor stability | Automatic — cascade protection kicks in under high CPU |
-| Config path respect | Set `sessions.claudePath` and `sessions.tmuxPath` in config.json |
-| iMessage mention fix | Automatic — 1:1 DMs always reach the agent |