feat: CLI-via-Goosed unified agent architecture with multi-agent routing by bioinfornatics · Pull Request #7238 · block/goose

bioinfornatics · 2026-02-16T00:15:04Z

CLI-via-Goosed: Unified Agent Architecture

Summary

This PR introduces a unified architecture where the CLI communicates with agents through goosed (the server binary), aligning desktop and CLI on a single communication path. It also adds multi-agent orchestration with an intent router, ACP/A2A protocol compatibility, and comprehensive UI improvements.

Key Changes

🏗️ Architecture: CLI-via-Goosed

CLI now communicates through goosed server instead of directly instantiating agents
GoosedClient manages server lifecycle (spawn, health check, graceful shutdown)
Process discovery & reuse via PID state file (~/.config/goose/goosed.state)
goose service install|uninstall|status|logs for managed daemon lifecycle (systemd/launchd)

🤖 Multi-Agent System

GooseAgent: 7 behavioral modes (assistant, specialist, recipe_maker, app_maker, app_iterator, judge, planner)
CodingAgent: 8 SDLC modes (pm, architect, backend, frontend, qa, security, sre, devsecops)
IntentRouter: Keyword-based routing with fuzzy prefix matching and configurable confidence thresholds
OrchestratorAgent: LLM-based meta-coordinator with fallback to IntentRouter
Internal modes (judge, planner, recipe_maker) filtered from public discovery

📡 Protocol Compatibility

ACP (Agent Communication Protocol): Full run lifecycle (create → stream → complete/cancel), elicitation, await flows
A2A (Agent-to-Agent): Dynamic agent card generation from IntentRouter slots
RunStore: Single-mutex design with LRU eviction (MAX_COMPLETED_RUNS=1000), TOCTOU-safe resume
ACP-IDE WebSocket: Session mode state with available/current mode tracking, notification forwarding

📊 Analytics & Observability

Routing analytics endpoints: POST /analytics/routing/inspect, POST /analytics/routing/eval, GET /analytics/routing/catalog
Routing evaluation framework: YAML-based test sets (29 cases), per-agent/mode accuracy metrics, confusion matrix
OpenTelemetry spans: orchestrator.route, orchestrator.llm_classify, intent_router.route
AgentEvent::PlanCreated: New event variant for orchestration plan tracking

🖥️ UI Improvements

WorkBlockIndicator: Collapsible tool-call chains with auto-open streaming, live-update panel
Progressive message rendering: Two-tier final answer detection, suppress transient tool call flash
Agent management: Dedup agents by ID, mode switching
ReasoningDetailPanel: Enhanced for work blocks with streaming support
Refactored hooks: useChatStream split into streamReducer.ts + streamDecoder.ts (860→576 lines)

🔒 Security & Reliability

Concurrency limit (10) on all /runs endpoints via ServiceBuilder
Structured ErrorResponse on all 11 bare StatusCode returns in runs.rs
ErrorResponse::bad_request() and conflict() constructors added
AcpIdeSessions LRU eviction (MAX_IDE_SESSIONS=100) with idle timeout

Quality Gates

Gate	Status
`cargo build --all-targets`	✅
`cargo fmt --check`	✅
`cargo clippy --all-targets -- -D warnings`	✅
`cargo test -p goose --lib` (789 tests)	✅
`cargo test -p goose-server` (40 tests)	✅
`npx tsc --noEmit`	✅
`npx vitest run` (325/326, 1 pre-existing)	✅
`npx eslint`	✅
Merge conflict check	✅ Clean

New Test Coverage

14 RunStore lifecycle tests: create/get, status transitions, await/elicitation, cancellation, events, output, errors, pagination, eviction
23 WorkBlock non-regression tests: streaming, completed, tool chains, final answer detection, dual indicator prevention
7 routing evaluation tests: YAML parsing, accuracy thresholds, metrics computation, report generation
6 SSE parser tests: event boundary handling, multi-event buffers, partial data
6 IntentRouter tests: keyword routing, fallback, disabled agents

Routing Evaluation Baseline

Overall: 41.4% (keyword router — LLM router pending)
Goose Agent:  100% (5/5)
Coding Agent:  33% (8/24)
Best modes:  architect 100%, qa 100%
Worst modes: frontend 0%, devsecops 0%, backend 20%

Files Changed

Rust: ~30 new/modified files across goose, goose-server, goose-cli, goose-mcp
TypeScript/React: ~15 new/modified files in ui/desktop
Tests: 50+ new tests (Rust + Vitest)
Docs: Architecture review, analytics backlog, protocol analysis

Follow-up Work

BL-2: Analytics UI dashboard (3-tab React page)
BL-3: Live user feedback (👍/👎 on routing)
Agent extraction: QA, PM, Security as standalone agents
LLM-based router (replace keyword matching)
Full OTel dashboard integration

michaelneale · 2026-02-16T00:50:12Z

thanks @bioinfornatics I like that general idea - looks like a lot of work to tidy up conflicts but would like to see what it looks like if you could show it here.

DOsinga · 2026-02-17T14:32:53Z

this is a massive change! I like the ideas, but I think we should discuss some of them separately (where we do we want to go with clients?) but also, it seems to introduce 5 big ideas, shouldn't we split those up?

…ution Implements the execution dispatch layer that bridges the orchestrator's routing decisions to actual execution backends: - Dispatcher trait: async dispatch_one/dispatch_all interface - InProcessDispatcher: local provider-based single-turn execution - A2ADispatcher: remote agent execution via A2A HTTP client - CompositeDispatcher: routes to correct backend based on DelegationStrategy - Concurrent fan-out for compound requests via futures::join_all - DispatchEvent broadcast for observability (started/progress/completed/failed) - Full test coverage for serialization, strategy validation, event variants Re-exports dispatch types through agents::orchestration facade module. Closes goose4-79bo

1. Fix Catalog/Workflows menu item swap — removed duplicate '/recipes' from Catalog zone, moved Apps from Workflows to Catalog zone 2. Apps no longer nested under Workflows — now properly in Catalog zone alongside Agents and Extensions 3. React DevTools: add fallback to load from local Chromium installation when electron-devtools-installer fails (known issue with Electron 40+) Sidebar zones after fix: - Workflows: Recipes, Scheduler - Catalog: Agents, Extensions, Apps - (removed duplicate Workflows entry from Catalog)

Add route: '/recipes' to Workflows zone so clicking the zone label navigates to the recipes overview (same UX as Catalog -> /catalogs). Both multi-item zones now behave identically: - Label click -> navigates to overview page - Chevron click -> toggles collapse/expand of sub-items

Adds a new GET /agents/catalog route that merges three agent sources into a single view for the Platform zone UI: 1. Builtin agents - from OrchestratorAgent slots with modes/capabilities 2. External ACP agents - from the ACP client manager 3. A2A instances - from the AgentPool with live status New types: CatalogAgent, CatalogAgentKind, CatalogAgentStatus, CatalogAgentMode, AgentCatalogResponse Closes goose4-9ib6

…t dropdown - Enhanced SessionList with always-on project grouping (≥2 projects or ≥5 sessions) - Added localStorage-backed project preferences (useProjectPreferences hook): - Pinnable projects (pinned always sort to top, visible pin icon) - Collapse state persistence across restarts - Recent project directories tracking (up to 10) - Added 'Open Project' dropdown button (FolderPlus icon) next to Chat: - 'Browse...' option opens OS directory picker via IPC - Recent project dirs listed below for quick access - Creates new session in selected directory - Home icon for 'General' project group (sessions without working_dir) - 10-project cap with search input when overflow - groupSessionsByProject now supports pin-aware sorting

- Add CatalogAgent, CatalogAgentKind, CatalogAgentStatus, CatalogAgentMode, AgentCatalogResponse to schemas - Add agent_catalog endpoint to paths - Regenerate frontend SDK types via just generate-openapi

…rs show parent context - Temp directories (/tmp/.tmp*, /tmp/tmp*) now grouped under 'General' instead of individual groups - Home directory sessions (no project context) grouped under 'General' - Nested subdirectories show parent context: 'goose4 › crates/goose' instead of just 'goose' - Verified with 278 real sessions: goose4 (101), General (96), goose4 › crates/goose (81)

… dir detection - General group now sorts first (was last) - General group always present even with 0 sessions - Temp dirs (/tmp/.tmp*, /tmp/tmp*) correctly grouped under General - Home dir detection uses regex patterns instead of unavailable appConfig keys - Nested subdirs show parent context: 'goose4 › crates/goose'

When OrchestratorAgent detects a compound request (is_compound=true with multiple sub-tasks), the reply route now: 1. Iterates over each sub-task in the plan 2. Reconfigures the agent for each sub-task's routing (apply_routing) 3. Executes each sub-task via agent.reply() sequentially 4. Aggregates results using aggregate_results() 5. Streams the combined result back as a single assistant message This closes the last gap between the orchestrator's routing intelligence and actual multi-agent execution. Single-task requests continue to use the existing direct reply path unchanged. Phase 1: sequential fan-out. Phase 2 will use AgentPool for parallel execution of independent sub-tasks.

…ete via API - Add Trash2 icon and onDeleteSession handler to SessionItem - Wire deleteSession API call with confirm dialog in AppSidebar - Navigate home if active session is deleted - Dispatch SESSION_DELETED event for sidebar state sync - Pass onDeleteSession through SessionList to flat and grouped views

- Add X button on project group headers (hover to reveal) - Click confirms then deletes all sessions in that project - Uses existing DELETE /sessions/{id} API for each session - Navigates home if active session was in the closed project - Dispatches SESSION_DELETED events for state sync - handleCloseProject wired through SessionList → AppSidebar

…er agent - Add SlotDelegation enum (InProcess, ExternalAcp, RemoteA2A{url}) - Add delegation_strategies field to AgentSlotRegistry - Add get_delegation, set_delegation, register_a2a_agent, register_acp_agent, unregister_agent methods - Wire register_acp_agent into connect_agent handler - Wire unregister_agent into disconnect_agent handler - Add 4 new tests covering delegation defaults, A2A registration, ACP registration, unregistration

- React Flow v12 (@xyflow/react) based visual pipeline editor - 7 node types: Trigger, Agent, Tool, Condition, Transform, Human, A2A - Drag-and-drop NodePalette with color-coded node kinds - PropertiesPanel with type-specific config forms per node kind - Undo/redo history (50-step cap) - YAML/JSON export via clipboard - Pipeline serialization (flowToPipeline, pipelineToFlow) - Route at /pipelines, sidebar entry under Workflows zone - Full TypeScript types for pipeline format (apiVersion goose/v1) - Semantic design tokens throughout, zero hardcoded colors

…oard, /active-agents, /health) - Add observatory.rs route module with 3 endpoints: - GET /observatory/dashboard — unified system health + active agents + performance snapshot - GET /observatory/active-agents — real-time agent status from pool + registry - GET /observatory/health — system health check with agent counts + status - Register observatory paths and types in OpenAPI schema - Generate frontend SDK with ObservatoryDashboard, SystemHealth, HealthStatus, ActiveAgent, ActiveAgentKind, ActiveAgentStatus, PerformanceSnapshot, TopTool types - Comment out broken pipeline utoipa references (frontend agent WIP)

- Add all_agent_names() to AgentSlotRegistry for enumerating all known agents - Replace hardcoded ['Goose Agent', 'Developer Agent'] sync in reply.rs with dynamic loop over all registered agents (builtin + external + A2A) - Comment out pipeline module/routes/openapi refs (frontend agent WIP, missing utoipa annotations — will be re-enabled when complete) - Ensures per-agent extension scoping works for dynamically registered agents

Backend (Rust): - crates/goose/src/pipeline.rs: Pipeline model, YAML/JSON serialization, validation with cycle detection, file-based CRUD storage - crates/goose-server/src/routes/pipeline.rs: REST API handlers for list/get/save/update/delete/validate pipelines with utoipa OpenAPI annotations - Registered pipeline module in lib.rs, routes/mod.rs, openapi.rs Frontend (TypeScript): - PipelineManager.tsx: Full pipeline list view with load/save/delete, wired to real API endpoints (listPipelines, savePipeline, getPipeline, deletePipeline) - Updated App.tsx route to use PipelineManager instead of standalone DagEditor - OpenAPI regenerated with all pipeline types and endpoints Pipeline storage: ~/.config/goose/pipelines/*.yaml

Backend (Rust): - crates/goose/src/pipeline.rs: Pipeline model with YAML/JSON serialization, validation with cycle detection, file-based CRUD in ~/.config/goose/pipelines/ - crates/goose-server/src/routes/pipeline.rs: REST API handlers for list/get/save/update/delete/validate with utoipa OpenAPI annotations - Registered pipeline module in lib.rs, routes enabled in mod.rs + openapi.rs Frontend (TypeScript): - PipelineManager.tsx: Pipeline list view with create/load/save/delete, wired to real API (listPipelines, savePipeline, getPipeline, deletePipeline) - Updated App.tsx route to use PipelineManager wrapper around DagEditor - OpenAPI regenerated with all pipeline types and 6 endpoints - Updated workflows/index.ts exports

Knowledge Graph: - +29 entities (total: 1,595), +36 relations (total: 2,258) - New: Pipeline system (backend + frontend), Session management, Generative UI integration, Design decisions, Findings Docs: - docs/non-obvious-knowledge.mdx: Hard-won insights (OpenAPI pipeline, React Flow compat, session grouping gotchas, design tokens, CSP warnings) - docs/reviews/ui-gap-analysis.mdx: Design vs implementation gap analysis - docs/reviews/ui-implementation-plan.mdx: 5-phase implementation plan - docs/reviews/project-sessions-assessment.mdx: Team assessment for project-grouped sessions feature

- Check SlotDelegation per sub-task in compound fan-out loop - RemoteA2A slots dispatch via A2AClient.send_message() over HTTP - InProcess/ExternalAcp slots continue using agent.reply() - Extract text from A2A Task artifacts or Message parts - Log delegation type per sub-task for observability - Completes the A2A dispatch loop: routing → delegation → execution

- Uncomment pub mod pipeline in routes/mod.rs - Uncomment pipeline::routes merge in configure() - Uncomment all pipeline path handlers in openapi.rs - Uncomment all pipeline schema types in openapi.rs - Add NodePosition + Viewport to OpenAPI schemas - Add instances.ts restore step to Justfile generate-openapi - These were repeatedly commented out by concurrent backend work

- Create WorkflowsOverview with 3 category cards: Recipes, Pipelines, Schedules - Each card loads real data from API (listSavedRecipes, listPipelines, listSchedules) - Shows status indicators (active/paused/draft), item counts, search, refresh - Update sidebar: Workflows zone route /recipes → /workflows - Add /workflows route in App.tsx - Pattern matches CatalogsOverview for consistent UX

…r cleanup

bioinfornatics force-pushed the feature/cli-via-goosed branch 6 times, most recently from e8dc59e to 4b68302 Compare February 17, 2026 12:59

bioinfornatics force-pushed the feature/cli-via-goosed branch from 059ea64 to 6604f1c Compare February 17, 2026 16:41

bioinfornatics added 21 commits February 17, 2026 19:13

feat(openapi): register agent catalog types in OpenAPI schema

1a56413

- Add CatalogAgent, CatalogAgentKind, CatalogAgentStatus, CatalogAgentMode, AgentCatalogResponse to schemas - Add agent_catalog endpoint to paths - Regenerate frontend SDK types via just generate-openapi

bioinfornatics added 22 commits February 22, 2026 13:19

genui: add structured jsonRenderSpec message parts

6416fe7

genui: allow unfenced json-render specs

c703749

docs: relax json-render fencing guidance

1c4a3d5

prompts: factor genui json-render output contract

de2a616

test: add genui contract consistency checks

456b623

ui: prefer shadcn json-render Table

5a5baa0

ui: fix Chart rendering in grid layouts

93f887a

orchestrator: route dashboards to genui

8a2a305

orchestrator: use dynamic catalog for mode selection

6d402b7

mcp: add genui service extension (sampling)

17fb0a1

config: enable genui_service extension by default

cf4bece

ui: guard json-render registry against shadcn overrides

eec757c

ui: export json-render component key lists

0728154

ui: harden json-render parsing (limits + unknown component errors)

f2e8bd5

ui: stabilize json-render setup exports

d6e2cac

ui: test json-render error surfaces

1ba2267

ui: validate json-render specs and improve error handling

8ff5f98

mcp: validate genui_service output types and size

225caca

routing: add dashboard intent tests

a767479

orchestrator: use slot registry for dynamic agent catalog

abecbf6

orchestrator: add DAG fields to plan tasks

6f2185f

Orchestrator: configurable max concurrency + compound DAG dispatcher

e14c65a

codefromthecrypt mentioned this pull request Feb 24, 2026

feat(claude-code): add permission prompt routing for approve mode #7420

Merged

3 tasks

bioinfornatics added 7 commits February 24, 2026 19:34

CLI: hermetic scenario tests via in-process goosed + fixture provider

0d4a2ea

cli: add per-invocation orchestrator max concurrency override

b57c9e6

orchestrator: persist compound run summaries to session extension data

9c6aaff

orchestrator: ingest compound runs into knowledge graph memory

46221eb

orchestrator: add dispatch event stream + run/task tracing spans

edbc5a9

ui: update GenUI example specs to match json-render registry

d28d0b1

ui: add regression tests for chat interruption queue and text animato…

7e6017f

…r cleanup

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Comments

feat: CLI-via-Goosed unified agent architecture with multi-agent routing#7238

feat: CLI-via-Goosed unified agent architecture with multi-agent routing#7238
bioinfornatics wants to merge 381 commits intoblock:mainfrom
bioinfornatics:feature/cli-via-goosed

bioinfornatics commented Feb 16, 2026

Uh oh!

michaelneale commented Feb 16, 2026

Uh oh!

DOsinga commented Feb 17, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Comments

Conversation

bioinfornatics commented Feb 16, 2026

CLI-via-Goosed: Unified Agent Architecture

Summary

Key Changes

🏗️ Architecture: CLI-via-Goosed

🤖 Multi-Agent System

📡 Protocol Compatibility

📊 Analytics & Observability

🖥️ UI Improvements

🔒 Security & Reliability

Quality Gates

New Test Coverage

Routing Evaluation Baseline

Files Changed

Follow-up Work

Uh oh!

michaelneale commented Feb 16, 2026

Uh oh!

DOsinga commented Feb 17, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants