[Bug] Agent file_write and file_delete report success when storage requests fail
Bewertung
- Schwierigkeit
- 3/5
- Geschätzter Aufwand
- 1-2 Tage
- Anfängerfreundlichkeit
- 76/100
- Issue-Typ
- Bug
- Klarheit
- Klar beschrieben
- Aktivitätsstatus
- Aktiv
- Tech-Stack
- typescript
Rechercherichtung
Start by reading executeFileTool and diskStorage.putTextFilesByJSON / deleteFilesByPaths, then trace how storage results become the Agent tool outcome. Add regression tests for HTTP 500 and rejected requests while preserving success behavior and best-effort behavior for other storage callers. Done means both file tools return an error for those failures, still return success for HTTP 200, and do not automatically retry.
Vom Indexierungsmodell aus dem Issue-Text verfasst.
Beschreibung
Describe the Bug
At main commit 02468154c4d99f8925916425bf444d672454fb3d, executeFileTool('file_write', ...) and executeFileTool('file_delete', ...) return success when the corresponding /api/session-data request returns HTTP 500 or rejects with a network error.
diskStorage.putTextFilesByJSON and deleteFilesByPaths do not check response.ok and swallow fetch rejections. The file-tool handlers therefore continue to their success result. ChatPanel adds that result to the next model request as the tool's outcome.
Steps to Reproduce
I exercised the unchanged exported executeFileTool through the real diskStorage implementation, replacing only fetch with synthetic responses. This is a Node-based module reproduction, not a browser or live-provider test.
For each of file_write and file_delete, use a synthetic path such as apps/diary/data/entries/test.json (and valid JSON content for the write), then test:
- fetch resolves with HTTP 500: actual result is
success - fetch rejects with
Error('synthetic offline'): actual result issuccess - fetch resolves with HTTP 200: actual result is
success, as expected
That gives four failure cases incorrectly reported as success and two passing controls. No real files or provider requests were used.
Expected Behavior
The Agent should receive an error result when storage returns a non-success HTTP status or the request fails. A network error does not prove that the server made no change, so this should not automatically retry the mutation.
Screenshots
No browser screenshots were captured for this module-level reproduction.
Environment
- OS: Linux
- Browser: not used for this reproduction
- Node.js: v24.19.0
- pnpm: not used for the original module probe
Additional Context
Proposed scope
Keep existing best-effort behavior for other storage callers. Let these two Agent tools opt into failure propagation, using the current error: ... result format. Preserve tool schemas, successful output, read/list behavior, paths, and batch scheduling. Concurrent batches can still partially complete; this is not an atomicity or rollback change.
I can contribute a small fix with real file-tool/storage-path regression tests. AI assistance was used for diagnosis and testing.
- Vorherrschende Sprache
- TypeScript
- Sterne
- 1.3k
- Forks
- 169
- PR-Merge-Kennzahlen
- Keine gemergten PRs in 30 T.
Entwicklungsumgebung
- Kein Dockerfile und keine Docker-Compose-Datei
- Hat eine Pull-Request-Vorlage
- Beitragsleitfaden lesen
Erste Schritte
- Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
- Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
- Forken Sie das Repository und arbeiten Sie in einem Branch.
- Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.
Mehr aus MiniMax-AI/OpenRoom
-
Schwierigkeit 5/5 Über eine Woche Anfängerfreundlichkeit 25/100
MiniMax-AI/OpenRoom#44 ·
Alle Issues in MiniMax-AI/OpenRoom
Ähnliche Issues
-
Schwierigkeit 1/5 Unter einer Stunde Anfängerfreundlichkeit 74/100
Maintainer antworten meist innerhalb von 2 Tagen
-
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 62/100
remotion-dev/remotion#12905 ·
Maintainer antworten meist innerhalb von 1 Tag
-
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 62/100
supermomonga/paseo-plugin-zcode-provider#83 · 1 Kommentar ·
Maintainer antworten meist innerhalb von 1 Tag
-
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 62/100
oxidecomputer/skepsis#102 · 1 Kommentar ·
Maintainer antworten meist innerhalb von 1 Tag
-
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 62/100
voidzero-dev/void#677 ·
Maintainer antworten meist innerhalb von 1 Tag