Adds _detect_verification_result() to parse QA agent stdout for failure
markers. Even if the QA agent exits with code 0, the merge will be blocked
if the output contains patterns like "verification failed", "requirements
not met", "not actually implemented", etc.
Also handles merge failures (e.g., conflicts) by moving to Triage.
Resolves: CG-62
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When verification succeeds but the merge to main fails (e.g., due to
conflicts or filesystem issues), the issue now moves to Triage state
for human intervention instead of continuing to Document state.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Dashboard:
- Parse tool_result blocks from user messages in stream view
- Display truncated tool results (max 500 chars) instead of "[User message]"
- Add tool_result styling with blue color and scrollable container
Backend:
- Fix null comment body handling in build_prompt (use `or ""` pattern)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Return "NO_CI" from _check_build_status for repos without build_type
- Add NO_CI handler in _process_build_items to skip to Verify state
- Add --verbose flag required for stream-json output format
- Sanitize task_id in history filenames (replace : with _)
- Remove agentrunner from build_types (no CI configured)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add secondary verification check that parses agent stdout for explicit
failure markers before merging to main. This prevents merging when the
QA agent reports "Verification Failed" in its output but exits with code 0.
Changes:
- Add _detect_verification_result() function to parse stdout for pass/fail
markers using regex patterns
- Modify _on_agent_complete() to check stdout content before calling
_merge_feature_branch()
- Log verification result (exit_code + stdout_check) before merge decision
- Block merge and add "Merge Blocked" comment when stdout indicates failure
- Move issue to Triage state when merge is blocked
Failure patterns detected:
- "verification failed"
- "requirements not met"
- "was not implemented" / "were not implemented"
- "not actually implemented"
- "none of the changes were implemented"
- "acceptance criteria not met"
- "## Verification Failed" header
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add AgentHistory class to persist completed agent runs to disk
- Store full message logs in JSON files with index for fast lookups
- Add /api/history and /api/history/{task_id} endpoints
- Create history page in dashboard showing past agent runs
- Add History link to navigation menu
- Update agent detail page to fall back to history for completed runs
- Display status badges (Success/Failed/Timeout) on history cards
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add StreamMessage dataclass and streaming output support to AgentTask
- Modify _start_agent to use --output-format stream-json for real-time output
- Add _read_output thread to parse streaming JSON and store messages
- Add get_task_output() method to AgentPool for polling output
- Add GET /api/agents/{task_id}/output endpoint for polling
- Add GET /api/agents/{task_id}/stream SSE endpoint for real-time streaming
- Create agent detail page at /agents/[taskId] with live output view
- Make agent cards clickable to view real-time output
- Add TypeScript types for streaming messages
Users can now click on a running agent in the dashboard to see
a real-time stream of the Claude Code conversation including
tool calls, responses, and results.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Fix PoolStatus.to_dict() field names to match frontend types
(active_count, active_tasks instead of active, tasks)
- Fix AgentTaskInfo.to_dict() to use start_time instead of started_at
- Fix _get_issues() to use correct YouTrackClient method signature
(get_issues_by_state takes project and state, not state and limit)
- Fix _get_builds() to use get_running_builds() instead of non-existent
get_builds() method
- Fix _get_issue_counts() to actually fetch counts from YouTrack API
instead of returning all zeros
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implements a full management dashboard for the Agent Runner at
https://agent.cleargrow.io with real-time monitoring and control.
Backend changes:
- Add oauth.py: Gitea OAuth2 authentication with session management
- Add api_server.py: REST API endpoints and static file serving
- Add dashboard_api.py: Data aggregation layer for dashboard
- Modify agent.py: Add kill_task() and get_task() methods
- Modify runner.py: Add event broadcasting and OAuth initialization
- Modify webhook_server.py: Integrate dashboard API handler
Frontend (SvelteKit + TypeScript):
- Dashboard overview with health status, agent pool, issue counts
- Agents page with active task list and kill functionality
- Issues page with state filtering and transitions
- Builds page with Woodpecker CI integration
- Config page for runtime settings
Features:
- Gitea OAuth2 login (same pattern as Woodpecker CI)
- Real-time status updates via polling (5s interval)
- Agent termination from dashboard
- Issue state transitions
- Service health monitoring
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The merge to main occurs after the QA verification step succeeds,
not after Review/Done.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
