Repository navigation

fix: Content-filter proxy + MCP servers for OpenCode on Databricks - #56

Merged

datasciencemonkey merged 21 commits into

mainfrom

fix/litellm-empty-content-blocks

Mar 11, 2026

Commits on Mar 11, 2026
- fix: add LiteLLM local proxy to sanitize empty content blocks for OpenCode
  
  datasciencemonkey
  and
  claude
  committed
  c0620bc
- docs: update design doc with clearer architecture description
  
  datasciencemonkey
  and
  claude
  committed
  42a78b7
- fix: remove invalid CLI flags from LiteLLM startup
  
  datasciencemonkey
  and
  claude
  committed
  cc7fb63
- fix: replace LiteLLM with lightweight custom proxy
  
  datasciencemonkey
  and
  claude
  committed
  ba49d30
- fix: comprehensive content-filter proxy with request + response fixes
  
  datasciencemonkey
  and
  claude
  committed
  517d21d
- fix: robust multi-pass orphan detection with diagnostic logging
  
  datasciencemonkey
  and
  claude
  committed
  b71a263
- fix: use @ai-sdk/openai for GPT models (Responses API support)
  
  datasciencemonkey
  and
  claude
  committed
  c31bf4e
- fix: explicitly install @ai-sdk/openai for GPT Responses API
  
  datasciencemonkey
  and
  claude
  committed
  228af71
- fix: rename LiteLLM references to content-filter proxy
  
  datasciencemonkey
  and
  claude
  committed
  f432f99
- fix: add Gemini compatibility — strip unsupported schema keys
  
  datasciencemonkey
  and
  claude
  committed
  01d8ad8
- fix: strip $schema from tool params for ALL models, not just Gemini
  
  datasciencemonkey
  and
  claude
  committed
  9f3bc1b
- fix: logging not writing — use explicit FileHandler
  
  datasciencemonkey
  and
  claude
  committed
  96ee510
- fix: logging to stderr (FileHandler buffering prevented writes)
  
  datasciencemonkey
  and
  claude
  committed
  c497445
- fix: kill stale proxy from previous deploy before starting new one
  
  datasciencemonkey
  and
  claude
  committed
  951e520
- fix: kill stale proxy by port instead of PID file
  
  datasciencemonkey
  and
  claude
  committed
  1684dc7
- fix: replace empty assistant content with placeholder instead of keeping
  
  datasciencemonkey
  and
  claude
  committed
  396dbb8
- feat: add DeepWiki and Exa MCP servers to OpenCode config
  
  datasciencemonkey
  and
  claude
  committed
  c023699
- refactor: remove all LiteLLM references
  
  datasciencemonkey
  and
  claude
  committed
  a03d981
- fix: update DeepWiki MCP URL from /sse to /mcp
  
  datasciencemonkey
  and
  claude
  committed
  d91aa17
- security: fail closed when app_owner can't be resolved (#57)
  
  datasciencemonkey
  and
  claude
  committed
  d1ad4e2
- security: add Content-Security-Policy header (#58)
  
  datasciencemonkey
  and
  claude
  committed
  30a9e4e

Navigation Menu

Repository navigation

fix: Content-filter proxy + MCP servers for OpenCode on Databricks - #56

fix: Content-filter proxy + MCP servers for OpenCode on Databricks#56
datasciencemonkey merged 21 commits into
mainfrom
fix/litellm-empty-content-blocks

Commits on Mar 11, 2026

fix: add LiteLLM local proxy to sanitize empty content blocks for OpenCode

docs: update design doc with clearer architecture description

fix: remove invalid CLI flags from LiteLLM startup

fix: replace LiteLLM with lightweight custom proxy

fix: comprehensive content-filter proxy with request + response fixes

fix: robust multi-pass orphan detection with diagnostic logging

fix: use @ai-sdk/openai for GPT models (Responses API support)

fix: explicitly install @ai-sdk/openai for GPT Responses API

fix: rename LiteLLM references to content-filter proxy

fix: add Gemini compatibility — strip unsupported schema keys

fix: strip $schema from tool params for ALL models, not just Gemini

fix: logging not writing — use explicit FileHandler

fix: logging to stderr (FileHandler buffering prevented writes)

fix: kill stale proxy from previous deploy before starting new one

fix: kill stale proxy by port instead of PID file

fix: replace empty assistant content with placeholder instead of keeping

feat: add DeepWiki and Exa MCP servers to OpenCode config

refactor: remove all LiteLLM references

fix: update DeepWiki MCP URL from /sse to /mcp

security: fail closed when app_owner can't be resolved (#57)

security: add Content-Security-Policy header (#58)