#19316: feat(agents): add agentic engineering workflows and /workflow command

## Summary This PR introduces **Agentic Engineering** as a practical workflow layer on top of OpenClaw subagents, with both UI and chat-command entrypoints. In simple terms: users can run proven multi-step engineering flows (research -> code -> review, bug triage, performance, security, etc.) using short-lived worker agents, then track/steer runs in one place. ## What Is Agentic Engineering? **Agentic Engineering** is a guided way to run software-engineering tasks as coordinated worker runs instead of one monolithic chat response. - Workers are **temporary subagents** spawned via `sessions_spawn` - Each step uses a role-appropriate template (`researcher`, `coder`, `reviewer`, etc.) - Runs remain inspectable and controllable with existing subagent controls - Persistent agent profiles remain unchanged; this is orchestration on top ## Why This Matters - Gives users repeatable execution flows for common engineering jobs - Improves quality by enforcing phased execution (analyze -> implement -> verify) - Makes long tasks safer and observable via run IDs + subagent controls - Reuses existing OpenClaw infrastructure (no new backend orchestration service) ## User-Facing Changes ### 1) Agentic Engineering workflow UX in Agents UI Added/expanded workflow-driven controls in the Agents view: - Dedicated **Agentic Engineering** section - Workflow selection and dispatch from UI - Clear visibility of available workflows Implemented workflows: - `research-code-review` (Research -> Code -> Review) - `bug-triage` - `refactor-safety` - `test-gap` - `performance-optimization` - `security-patch` ### 2) Built-in worker templates expanded Subagent templates now include: - `researcher` - `coder` - `reviewer` - `bug-triager` - `reproducer` - `test-builder` - `profiler` - `security-researcher` ### 3) New chat/native command: `/workflow` Added command support for workflow operations from chat/TUI/native command surfaces: - `/workflow list` - `/workflow run <workflow-id> <goal>` - `/workflow run <workflow-id> [--label <prefix>] [--timeout <seconds>] [--cleanup keep|delete] <goal>` - `/workflow status` Behavior: - `/workflow run ...` dispatches a workflow orchestrator run via the current session - Execution uses `sessions_spawn` step-by-step with carry-forward instructions - Progress/child runs remain manageable through `/subagents ...` ## How To Use ### UI path 1. Open `Agents` 2. Open `Agentic Engineering` 3. Select workflow + enter goal 4. Run workflow 5. Monitor and control via chat `/subagents ...` ### Chat path - List flows: `/workflow list` - Run a flow: `/workflow run bug-triage "Investigate flaky auth-login test"` - Check status summary: `/workflow status` - Inspect full run details: `/subagents list` ## Implementation Notes - Workflow orchestration prompt generation aligns with Agentic Engineering UX expectations - Dispatch is non-blocking and compatible with existing agent lifecycle APIs - Command is authorization-gated like other management commands - No destructive changes to existing agent profile model ## Documentation Updates Updated docs to reflect current capabilities: - Added `/workflow` command to slash command docs - Updated subagent/session docs to include expanded built-in templates ## Validation / Testing Executed locally: - `pnpm tsgo` - `pnpm lint src/auto-reply/reply/commands-workflow.ts src/auto-reply/reply/commands-core.ts src/auto-reply/commands-registry.data.ts src/auto-reply/reply/commands.test.ts` - `pnpm vitest --config vitest.unit.config.ts src/auto-reply/reply/commands.test.ts` - `pnpm lint:docs` Additionally validated behavior manually in local UI and chat: - Workflow controls visible in Agents > Agentic Engineering - Workflow dispatch executes and produces subagent runs - `/subagents` remains the run-level control surface ## Backward Compatibility - Existing `/subagents` and `sessions_spawn` behavior is preserved - Existing agent profiles and routing remain unchanged - Feature is additive and opt-in <!-- greptile_comment --> <h3>Greptile Summary</h3> Introduces Agentic Engineering workflows with `/workflow` command and `mesh` plugin extension for orchestrated multi-step software engineering tasks. Key additions: - `/workflow` command supports list, run, and status operations for predefined workflows (research-code-review, bug-triage, refactor-safety, test-gap, performance-optimization, security-patch) - UI panel in Agents view provides workflow selection and dispatch controls - `mesh` plugin adds lower-level workflow orchestration with auto-planning, execution, retry, and status tracking via gateway methods - Comprehensive test coverage for workflow command parsing and execution flows - Documentation updated to reflect new command availability <h3>Confidence Score: 4/5</h3> - Safe to merge with minor observations - implementation is solid with good test coverage and proper authorization checks - Code is well-structured with proper validation, authorization gating, and comprehensive tests. The mesh plugin introduces significant new functionality but follows established patterns. Minor style inconsistencies and potential edge cases exist but don't block merge. - Pay attention to `extensions/mesh/src/mesh-gateway.ts` for the core workflow execution logic and ensure timeout values are appropriate for production workloads _{Last reviewed commit: 0e081e5} <!-- greptile_other_comments_section --> _{(4/5) You can add custom instructions or style guidelines for the agent [here](https://app.greptile.com/review/github)!} <!-- /greptile_comment -->