* Add two-phase tool call approval for agents in channels
Implements the mobile counterpart to the webapp's multiplayer tool calling
feature. When a bot is @mentioned in a channel, tool call arguments and
results are redacted from other members. Only the invoker can approve/reject
tool execution (Phase 1) and decide whether to share results with the
channel (Phase 2).
See mattermost/mattermost-plugin-agents#491 for the server/webapp changes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Add tests for channel tool calling utilities and remote actions
Tests for isToolCallRedacted, isPendingToolResult, getToolApprovalStage,
mergeToolCalls utility functions and fetchToolCallPrivate,
fetchToolResultPrivate, submitToolResult remote actions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Add e2e tests and testIDs for agent tool calls in channels
Adds detox e2e tests covering tool call card rendering, approval buttons,
result approval phase, and multi-tool-call scenarios. Adds testID props
to ToolApprovalSet and ToolCard components to support the e2e tests.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Address PR review feedback for two-phase tool call approval
- Fix mergeToolCalls to preserve public-only tools instead of dropping them
- Fix stale closure race in handleToolDecision with functional setState
- Add forceLogoutIfNecessary to tool_private and tool_result remote actions
- Show snackbar on submit failure using existing error types
- Reset isDM in catch block to prevent stale state
- Sync animation shared values when isCollapsed changes externally
- Clear private data on streaming-to-persisted transition
- Wrap action buttons with usePreventDoubleTap
- Use toolCalls reference instead of toolCalls.length in effect dependency
- Fix grammar in warning callout ("its" -> "their")
- Change fontWeight from number to string per RN conventions
Co-authored-by: Cursor <cursoragent@cursor.com>
* Address PR review feedback: withObservables HOC, memoization, typography
- Refactor AgentPost to use withObservables HOC for channel observation
instead of useEffect+subscribe, providing isDM as a prop
- Memoize undecidedCount in ToolApprovalSet and move before early return
- Replace manual font styles with typography() utility in ToolCard
Co-authored-by: Cursor <cursoragent@cursor.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
* Agents streaming
* Fix bug and lint
* Add reasoning summaries to mobile agents
* Add tool call approval UI for mobile agents
* Add citations and annotations support for mobile agents
Implements Phase 4 from the agents mobile plan:
- WebSocket event handling for annotations control signal
- CitationsList component displaying sources below agent responses
- Collapsible sources section with source count
- Each citation shows title, domain, and opens URL on tap
- Citations persisted in post.props.annotations
- Touch-optimized UI with 44pt minimum tap targets
* Add stop and regenerate controls for mobile agents
* Add tests
* Fix tool approval buttons not updating after accept/reject
- Remove optimistic WebSocket event approach that didn't work
- Clear component streaming state on ENDED event to force switch to persisted data
- Remove delays in streaming store cleanup (500ms and 100ms)
- POST_EDITED event now properly updates UI for both accept and reject
- Remove debug console.log statements
- Fix ESLint issues (unused vars, nested callbacks, nested ternary)
* Refactor agents code: remove unused client mixin, fix bugs, and simplify logic
* Fixes
* Add CLAUDE.md
* Review feedback.
* Review feedback inline functions
* Learnings
* Address review feedback: StyleSheet.create, parent mount checks, utils
* Move to observables
* Last style fix
* Style tweaks
