truefoundry/trueforge @truefoundry/trueforge@0.180.0

on GitHub

Patch Changes

• 36a0bfa: Preserve supervisord ENTRYPOINT on Daytona custom-environment buildInfo snapshots so the sandbox NATS bridge starts (Daytona otherwise defaults to sleep infinity).
• d233f44: Set Daytona requestTimeoutMs to 60s so the SDK applies a per-exec HTTP deadline (command timeout + 5s) and a hung toolbox response cannot leave a turn blocked without a tool.response.
• 0a1606d: Default the reasoning effort to medium when the selected model offers it, instead of defaulting to the lowest effort. TrueFoundry model listing now passes mergeParams to the provider-integrations API so provider-level parameter defaults are included in each model's advertised parameters.
• 0a42995: In TrueFoundry mode, x-tfg-skills, x-tfg-mcp, and x-tfg-models are allowed when creating an inline session, replacing its spec, or running a turn on it. Create and replace only use the headers to check the spec and do not save them. x-tfg-skills must list every skill as a git skill; registry skill ids are not accepted. MCP servers and models fall back to the registry when a name is missing. Other routes reject these headers.
• 3b2405e: Add tenant, request, session, and turn ids to server and controller logs so a failure can be filtered to one request or turn. Log lines no longer include the package version.
• ce2a843: Add newer model presets to the shipped model catalog for the providers it already lists.
• 26c77c4: Add optional AgentSpec-shaped model ({ name: provider/model, params? }) on model.message events, stamp it from the bound agent definition for new turns, and backfill the session tip turn's events / turn output from the session agent model FQN.
• 8cd3f33: Expose Prometheus request metrics at GET /metrics so a scraper can see rate, status, and latency for /api requests.
• f10e3b8: Default reasoning effort now follows medium → high → low in both the draft (New Chat) composer and the agent builder (Agent Config) when the selected model does not offer medium, instead of falling back to the lowest offered effort.
• d3d80a2: Use the rediss:// URL scheme for standalone Redis when REDIS_TLS_ENABLED is true so TLS connections advertise the secure protocol in the connection URL.
• b037c92: Only validate model.params.reasoning_effort against a model's advertised efforts when that list is non-empty, so models without configured efforts no longer reject specs that set it.
• b575b9a: Compare the service API key with crypto.timingSafeEqual so a rejected credential does not leak how much of the key matched.
• 9208cc8: Allow tenant members to use sandbox environments created by other members while keeping update and delete actions restricted to the creator.
• 5da156a: Add a host-overridable Register Skills action next to the Skills modal title. The default slot waits for capabilities, routes to Settings → Skills or a platform URL, and hides when neither applies.
• eb5560c: Add a keyless You.com remote MCP server preset to the shipped MCP catalog so users can opt into You.com web search alongside the existing presets.
• Updated dependencies [5dd0089]
• Updated dependencies [36a0bfa]
• Updated dependencies [0a1606d]
• Updated dependencies [5fd9cca]
• Updated dependencies [26c77c4]
• Updated dependencies [be2bb87]
• Updated dependencies [248e768]
• Updated dependencies [05cef0b]
  • @truefoundry/trueforge-core@0.180.0
  • @truefoundry/trueforge-sdk@0.180.0