Spaces:
Running
Running
Commit History
30 days ago
-
Sync backend to GitHub f8d0b96: consent-gated memory, analytics API, structured logging
siddhm11Brings the Space up to date with 9 backend commits from GitHub main (ad58940..f8d0b96): - Passive memory written only on /enhance/accept, retrieval filtered on approved=true; existing unapproved vectors stop matching - Retrieval thresholds calibrated per layer (saved 0.24, passive 0.20) - New /api/user/analytics, /api/evaluate/prompt-improvement, /api/user/prompt-history/{id} - Account deletion ordered, retryable, profile deleted last - Prompt text removed from console logs; print -> JSON logger - Embedding LRU cache shared across memory layers Dockerfile, README.md and other Space-only files unchanged.10ce33e
-
Sync backend to GitHub d419447: voice transcription, builder dashboard, analytics
siddhm11 +1: Claude Opus 5Brings the Space up to date with siddhm11/prompt_engineering_skeleton@d419447. Adds POST /voice-transcribe (the extension has been calling it and getting 404), the private builder dashboard API, and analytics telemetry. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NoNHrDfqGj7BLKagHkRM3Zb1f9886
-
Harden the vector store against the failure modes that hid a real outage
siddhm11 +1: Claude Opus 5Connectivity is now verified before a client is accepted; provisioning latches only on success and is retried; a broken client can be reset; reconnection is rate-limited; dimension mismatches are refused loudly; requests carry a timeout; and failed reads no longer look identical to empty results. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015akmAZh7sN51giQFyMuDKRc61e3b2 -
Report Qdrant URL shape so a bad config is distinguishable from a dead cluster
siddhm11 +1: Claude Opus 5Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015akmAZh7sN51giQFyMuDKR0352dbc -
Report vector-store health alongside embedding health
siddhm11 +1: Claude Opus 5Point counts are the decisive signal for why saved-prompt search matches nothing: empty means writes never landed, populated means the query is at fault. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015akmAZh7sN51giQFyMuDKR72bd44e -
Deploy the audit remediation backend
siddhm11 +1: Claude Opus 5Brings the Space in line with the tested tree. Two changes here are what production is currently missing, and sign-in cannot work without both: - /auth/google/login now returns a `state` handle - /auth/google/poll exists at all (it currently 404s) The extension's rewritten sign-in polls that endpoint from the service worker, because an MV3 action popup is destroyed the moment it loses focus and took the old postMessage listener down with it before Google ever redirected back. Also lands: the daily ration no longer fails open when Mongo is unreachable (one read failure previously granted everyone unlimited enhancements, which can drain the shared Groq org quota in minutes); Qdrant point ids are deterministic so deleting a saved prompt actually removes its vector; per-user rate limiting and a request body cap are wired up; CORS no longer pairs allow-all with credentials and /docs is closed in production; pasted credentials are redacted before being logged or embedded; and /voice-e…bd719d5
-
Update backend from polished-with-website branch - new routers (feedback, saved_prompts), updated services and configs
siddhm11c0dd1c5
-
Update backend: mode-aware enhance, saved prompts CRUD, voice-enhance, feedback endpoint
siddhm114d5cbc7
-
changes in sim .9 , database:
siddhm11d3974e0
-
refactor for HF spaces: moved code to backend/
siddhm11515a3fb