deepspace 0.28.2 → 0.29.1

package/CHANGELOG.md

@@ -1,5 +1,29 @@
 # deepspace
 
+## 0.29.1
+
+### Patch Changes
+
+- Harden the integration catalog fetch used by `deepspace integrations
+list/info/invoke`: the request is now bounded by a 15-second timeout, and an
+  unreachable service, invalid JSON, or a wrong response envelope surfaces as a
+  clean `catalog_unavailable` / `invalid_catalog` refusal instead of a raw
+  network or parse error. One shared fetcher now serves the CLI and the
+  repo-only integration-health scanner.
+
+## 0.29.0
+
+### Minor Changes
+
+- Add `logEventText(event)` to the log-events module: one event's plain-text body (what the event says, without colors, tags, timestamps, or stack frames), shared by the CLI line formatter and the dashboard log rows and search. CLI output is byte-identical; this is the extraction of the wording both renderers already produced independently. The platform logs endpoint also retires its never-used `until` query param — the window always ends at request time, and an explicit `until` is ignored (no released client ever sent one).
+- New `loggableError(err)` export (shared, both entries): renders an error as the string form that survives Workers Logs ingestion — the telemetry pipeline keeps only a logged Error OBJECT's stack frames, dropping its message, so uncaught route errors showed in `deepspace logs` as bare frames. The SDK's own error logs now go through it — the room runtime (BaseRoom/RecordRoom/CronRoom/JobRoom message, task, job, and websocket error logs), Yjs doc loads, the auth glue's Apple-secret/onUserCreated logs, chat compaction, the files quota/multipart paths, and the documentation MCP worker — so those messages come back for every app on upgrade + redeploy. Output is bounded: causes are walked to a fixed depth and the rendered text is capped at the exported `MAX_LOG_TEXT_LENGTH` (8 KB, the same budget the platform's log reader truncates at), so a hostile or enormous error can't flood the log store. App code should follow the same rule: log `loggableError(err)`, never a bare Error object.
+- Remove the opt-in client-side error forwarding subsystem (`installClientErrorReporter`, `reportClientError`, `registerClientErrorRoute`, `handleClientErrorReport`, `CLIENT_LOG_MARKER`, and the `ClientErrorReport`/`ClientErrorKind`/`ClientErrorReporterOptions` types), and with it the `source: 'client'` field on `AppLogEvent` and the `CLIENT` tag in `deepspace logs` and the dashboard. Shipped in 0.7.0, it was never wired anywhere — not the scaffold, the skill, or the site manual — so it has been ~900 lines of dormant surface, including an anonymous ingestion endpoint (`POST /_deepspace/client-errors`) in every app that opted in. Browser JS errors are therefore invisible to `deepspace logs` again (they never invoke the Worker); the tenant-safe design is recorded in `docs/platform/app-logs.md` and observability gap A7 for resurrection — scaffold-wired by default — if demand appears. Apps that did call the removed APIs must drop those calls when upgrading; reports POSTed by already-deployed apps now surface as plain marker-prefixed log lines instead of `CLIENT`-tagged events.
+- New `workerErrorHandler(prefix, respond?)` export (`deepspace/worker`): the one Hono `app.onError` — a response-bearing error (`HTTPException`) keeps its own answer, everything else is logged as one string (`[prefix] METHOD /path: message, frames, bounded cause chain`, via `loggableError`) and answered with a generic 500 (or the worker's own `respond` shape). The scaffold template and the deploy/auth/platform workers now register it instead of carrying hand-copied handler bodies (api-worker's own handler is the billing owner's area, unchanged). Also exported: `truncateLogText(s)`, the single budget-inclusive, surrogate-safe truncation both `loggableError` (write side) and the platform log reader (read side) use — a capped field never exceeds `MAX_LOG_TEXT_LENGTH` anymore. And `--search` now matches the shared `logEventText` rendering (the same corpus the dashboard searches and both renderers display), so an exception's name/message and a request's method/path/status match — not just the raw message field.
+
+### Patch Changes
+
+- `logs --follow` no longer retries a 400 forever. The flags are constant for the life of a tail (a bad `--since` already failed on the first fetch, before the loop), so a mid-tail 400 means the platform stopped accepting a shape it used to — exactly what removing a param does to an older CLI. Resending the identical request cannot succeed; the tail now ends with the error. 429 (the route's own throttle) and 5xx still back off and retry.
+
 ## 0.28.2
 
 ### Patch Changes

package/dist/{chunk-DMUR67NG.js → chunk-DWSZYN73.js}

@@ -1274,7 +1274,7 @@ function renderInlineMarkdown(value) {
 import { useEffect as useEffect6, useRef as useRef5, useState as useState5 } from "react";
 
 // package.json
-var version = "0.28.2";
+var version = "0.29.1";
 
 // src/documentation/runtime/navigation.tsx
 import { Fragment as Fragment4, jsx as jsx6, jsxs as jsxs6 } from "react/jsx-runtime";
@@ -1954,4 +1954,4 @@ export {
   DefaultDocumentation,
   DocumentationContent
 };
-//# sourceMappingURL=chunk-DMUR67NG.js.map
+//# sourceMappingURL=chunk-DWSZYN73.js.map