-
4.16 tmp (#7404) · 9bb6b82c* Agent sandox user (#7383) * feat(sandbox): implement user-level sandbox and legacy workspace migration (#7315) * feat(sandbox): implement user-level sandbox and legacy workspace migration * refactor(sandbox): implement durable-saga pattern and migrate lifecycle operations * Revert "refactor(sandbox): implement durable-saga pattern and migrate lifecycle operations" This reverts commit 921953d97819776645868038e44d75decd552e8a. * refactor(sandbox): implement durable-saga runner and migrate lifecycle operations * fix(sandbox): preserve legacy migration backups * fix(sandbox): repair missing provider instances and support allowCreate option * refactor(sandbox): cleanup legacy archive endpoints and streamline migration workflow * feat: add direct sandbox workspace preview (#7316) * feat: add direct sandbox workspace preview * refactor: align sandbox preview claims with upstream refactor: fix IDE agent internal ports fix: decode sandbox proxy claims generically ci: install Rust toolchain without third-party actions * chore: keep sandbox proxy on one port by default feat: split sandbox proxy preview port fix: require sandbox preview proxy URL * Codex/agent sandbox preview (#7349) * refactor: use sessions for sandbox preview links * docs: update sandbox preview session design * fix: align preview sessions with user sandboxes * fix: align preview ids with sandbox runtime * refactor(sandbox): unify App and Skill runtime image upgrade status (#7352) * refactor(sandbox): simplify provider migration and runtime upgrade logic (#7357) * feat(sandbox): improve app chat runtime resilience (#7361) * refactor(sandbox): implement server-side silent runtime upgrade for App workflows * feat(sandbox): implement graceful degradation for app chat when sandbox is unavailable * refactor(sandbox): refactor ticket verification and rename preview session cache * fix(sandbox): pause and reuse OpenSandbox resources (#7363) * fix(sandbox): pause OpenSandbox resources on stop * refactor(sandbox): upgrade OpenSandboxAdapter lifecycle and SDK integration * refactor(sandbox): align adapter contract, structure and OpenSandbox lifecycle * feat(sandbox): make suspend and archive thresholds configurable via env * refactor(sandbox): unify workflow sandbox preparation and status streaming * fix(sandbox): create parent directories before file writes (#7384) * feat(sandbox): prepareSandboxFileParentDirectories utility and update sandbox file handling * doc: summary * feat(sandbox): improve resource controls and agent reliability * doc * doc * fix(sandbox): move migration staging out of workspace (#7394) * fix(sandbox): move migration staging out of workspace * refactor(sandbox): replace batchRun waitForAll with batchRunSettled * env * fix(sandbox): restore legacy migration prerequisites (#7396) * fix(sandbox): restore legacy migration prerequisites * fix(sandbox): stop migrated targets * fix(sandbox): harden preview sessions and document risks * fix(sandbox): add preview-only proxy env validation * docs(sandbox): clarify required preview proxy env --------- Co-authored-by: DigHuang <114602213+DigHuang@users.noreply.github.com> * feat(workflow): support agent generated tool parameters (#7267) * feat(workflow): add agentGenerated input type for tool parameters - Add `FlowNodeInputTypeEnum.agentGenerated` to distinguish parameters generated by the model during tool calls. - Implement `initToolInputTypeByDefaultMode` to handle initialization based on tool descriptions and safe boundaries. - Update ToolCall and Agent runtime schema generators to filter parameters based on final input type. - Add security boundaries to prevent sensitive fields (e.g., secrets, files, model selection) from being marked as agent-generated. - Update UI components to support configuration of parameter sources and improve state management for tool inputs. * fix(workflow): surface agent generated tool params * feat(workflow): align agent generated tool input UI * refactor: simplify plugin runtime and cleanup workflow logic * feat(workflow): introduce isToolParam to define agent-generated inputs Add `isToolParam` to handle default input generation logic more precisely than `toolDescription`, ensuring better control over which fields are exposed as model parameters in the tool call schema. * fix(workflow): honor selected render type index * Restrict agent tool params to selected generated fields * Persist selected tool input type in workflow detail * fix(formEdit): preserve saved input type when initializing defaults * fix(workflow): avoid materializing unset selectedType * refactor(formEdit): improve tool input selection * refactor(formEdit): drop default tool input mode Remove persisting default tool input mode (isToolParam) and centralize selection via saved input by introducing stripToolInputDefaultMode and updating call sites to pass savedInput only. Propagate tool errors through the agent loop by carrying errorMessage in tool results and dispatch, so errors do not produce visible node responses. Update tests, runtime types, and i18n descriptions accordingly. * fix(form): normalize reference-only tool inputs to agent generated * feat(global): preserve workflow tool node metadata in json schema * refactor(config-tool-modal): unify input and output sections with shared * fix(formEdit): default user chat inputs to agent generated in tools * fix(tool): align runtime schemas with legacy tool inputs * fix(jsonschema): preserve open union types and secret config state * refactor: improve application workflows and supporting modules * fix(form-edit): preserve textarea render type selection * fix(workflow-tool): preserve hidden inputs and normalize legacy defaults * fix(workflow): migrate legacy tool descriptions to AI inputs * fix(workflow): scope legacy tool input migration to supported tools --no-edit * fix(workflow): restore legacy workflow tool input types at runtime * fix(app): preserve legacy tool input compatibility * fix(workflow): restrict legacy tool description fallback * feat(agent): persist tool input modes independently * refactor(tool): move agent tool schemas out of skill module * fix(tool): filter agent-generated params from tool configs * refactor: reorganize application modules and shared services * fix(workflow): align tool input defaults and option handling * fix(form-edit): preserve explicit input types and infer tool controls * feat(agent-tools): support fixed and latest tool versions * chore: submodule * fix(workflow): preserve fixed tool versions and legacy input types * submodule * fix(sandbox): align v2 migration and Devbox resource limits (#7405) * feat: clean dead code * refactor(service): remove tmbId & legacy fields from skill runtime APIs * feat(sandbox): pass resource limits to Sealos Devbox * refactor(sandbox): share resource limit configuration (#7407) * fix(workflow): preserve tool node input configuration (#7409) * fix(workflow): align tool input defaults and option handling * fix(workflow): preserve tool node input configuration * doc * doc * fix(workflow): restore tool parameter label (#7410) --------- Co-authored-by: DigHuang <114602213+DigHuang@users.noreply.github.com> Co-authored-by: Finley Ge <32237950+FinleyGe@users.noreply.github.com>
Archer committed
