feat(agent): allow variable tool permission modes - #7538
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub. |
|
@cubic-dev-ai review this PR |
@j15z I have started the AI code review. It will take a few minutes to complete. |
|
11c3d58 to
818dca4
Compare
818dca4 to
41af1d3
Compare
41af1d3 to
661f655
Compare
|
|
@cubic-dev-ai review this PR |
@j15z I have started the AI code review. It will take a few minutes to complete. |
|
@cubic-dev-ai review this PR |
|
@cubic-dev-ai review this PR |
@j15z I have started the AI code review. It will take a few minutes to complete. |
|
@cubic-dev-ai review this PR |
@j15z I have started the AI code review. It will take a few minutes to complete. |
|
@cubic-dev-ai review this PR |
@j15z I have started the AI code review. It will take a few minutes to complete. |
|
@cubic-dev-ai review this PR |
@j15z I have started the AI code review. It will take a few minutes to complete. |
Preserve the new MCP operation policy editor and runtime pipeline alongside canonical tool permissions. Keep structured response-read failures and interrupted-write guidance together in the CLI.
|
@cubic-dev-ai review this PR |
@j15z I have started the AI code review. It will take a few minutes to complete. |
Allow one flush per subblock while newer edits coalesce in a separate pending batch. Start the next ready batch after the active save completes, including failures, so a slow older lookup cannot overwrite a newer edit. Cover delayed lookups, coalescing, failure recovery, confirmation order, and independent subblock saves with regression tests.
|
@cubic-dev-ai review this PR |
@j15z I have started the AI code review. It will take a few minutes to complete. |
Summary
<start.toolMode>. Resolve variables before filtering tools and reject invalid modes before calling the provider.Companion: simstudioai/mothership#488
Type of Change
Testing
Remaining coverage: the latest staging merge has automated coverage but was not rerun in Chrome. Undo/redo, additional collaborative/network interleavings, viewer-account behavior, live LLM-provider outage injection, global cron ticks, and queue transport remain unverified. The existing multi-request importer is not transactional for arbitrary network failures.
Review focus: keeping each tool's permission attached through array edits, excluding disabled tools from discovery/execution, and preserving both selector and variable inputs across workflow operations.
Checklist
Screenshots/Videos
No screenshots or videos attached.