Feature/multi agent taskid collab 20251114

.askman/tasks/20251114-2255-rebuild-multi-agent-comm-demo/README.md

@@ -0,0 +1,59 @@
+# Rebuild Multi-Agent Communication (Quick Dev Flow Demo)
+
+## Requirement (2025-11-14T16:13:10.912Z)
+- The Feature-Plan stage MUST support interrupt/abort when it detects conflicts with existing requirements.
+- On interrupt, the sequential pipeline MUST short-circuit and emit a structured failure payload (e.g., { ok: false, code: 'CONFLICT', message, details }) or a controlled abort that the orchestrator maps to a failure, preventing downstream stages (feature-edit, YAML validator, report).
+
+- Task ID: 20251114-2255-rebuild-multi-agent-comm-demo
+- Goal: Rebuild/streamline multi-agent communication (DevHub ↔ LoopAgent/SequentialAgent) for a quick development flow demo.
+- Notes:
+  - Focus on task-id centric collaboration across agents
+  - Keep CLI UX stable; prefer npx usage
+  - Prepare minimal demo path via DevHub + Blueprint + Coder/Review
+
+## Additional Notes (2025-11-14T15:00:15.263Z)
+- Atomic agent types: Prompt Agent and Programming Agent
+- Prompt Agent is prompt-driven, focusing on a specific task type's working method
+- Programming Agent can have multiple nodes; nodes are connected via program logic, and can be defined so each node can be emitted independently
+
+## Design Target & Use Case (2025-11-14T15:08:00.563Z)
+- Agent design is centered on artifacts + working method, not just raw prompts.
+- Use Case: feature-updater agent
+  - Goal: After edits, ensure features.yaml remains well-structured.
+  - Validate: No duplicate or conflicting feature entries (避免重复/冲突).
+  - Problem with pure Prompt Agent: LLM can introduce subtle structural mistakes that propagate downstream.
+  - Decision: Use Programming Agent architecture (multi-node, logic-enforced) to guarantee deterministic validation & merge rules before committing updates.
+- Rationale: Programming Agent nodes can emit structured events (parse, diff, validate, normalize) ensuring post-edit artifact integrity and reducing cascading errors in later workflow stages.
+
+## Code Review By Features Agent (2025-11-14T15:40:26.943Z)
+- Purpose: Verify whether new code changes (commit diffs) risk breaking previously implemented features.
+- Context: When adding new features quickly, we validate the new one but often skip deep regression of old ones.
+- Approach: Iterate existing features (from features.yaml or feature artifacts) and cross-check each feature's logic paths against changed code segments.
+- Detection: Identify overlaps (cross points) where modified functions, modules, or data contracts intersect a feature's required invariants.
+- Output: Structured report listing per-feature risk status, with:
+  - feature_id / name
+  - risk_level (none/low/medium/high)
+  - impacted elements (files, functions, lines)
+  - rationale (what change introduces the potential break)
+- Architecture Choice: Programming Agent (multi-node) for deterministic traversal: nodes = [collect_diff, enumerate_features, map_dependencies, detect_cross, score_risk, generate_report].
+- Benefit: Early surfacing of silent regressions without full test suite expansion.
+
+## Planned Upgrade: Feature-Updater Sequential Agent (2025-11-14T15:59:12.767Z)
+- Goal: Upgrade to feature-updater with a sequential architecture; rename feature-writer to feature-edit.
+- Composition:
+  - Feature-Plan (Prompt Agent; new agent)
+  - Loop (LoopAgent; current: blueprint loop)
+    - Stage: feature-edit (Prompt Agent; renamed from feature-writer; supports add/delete/change) → YAML validator
+  - Feature-Changes Review (Prompt Agent) to check and produce a report
+- Pipeline: feature-plan → loop(feature-edit → yaml validator) → feature-changes report.
+- Rationale: Combine prompt-driven ideation with programmatic validation to reduce LLM-induced structural errors while producing a final human-readable risk/change report.
+
+> 2025-11-14T16:00:41.950Z: Note — feature-edit is a Prompt Agent.
+
+## Additional Agent Evolution (2025-11-14T16:21:31.043Z)
+- Blueprint will become a Prompt Agent (was LoopAgent) focused on direct user dialogue and selectively invoking other agents when needed (not a full orchestrator/central workflow controller).
+- Selection logic: interpret user intent → choose Feature-Updater, UI Review, Notes Agent, or others; skip calls when no added value.
+- UI Review (ui-review) remains a Prompt Agent for UI/UX; legacy structure requires audit/refactor.
+- Notes Agent (new Prompt Agent): writes fragments to notes-<timestamp>.md in task workspace; appends decisions, conflicts, risks chronologically.
+- Example interaction path: user intent → blueprint (dialog) → conditional calls: (feature-updater loop) and/or (ui-review) → notes-agent summary.
+- Abort policy: On conflict detected in feature-plan or validation failure, blueprint stops further calls and invokes notes-agent to persist the issue.

.gitignore

@@ -33,6 +33,7 @@ coverage/
 
 # Bundlers and frameworks
 .parcel-cache/
+.vite-cache/
 .next/
 .nuxt/
 .svelte-kit/

package.json

@@ -36,6 +36,8 @@
     "tsup": "^8.5.1",
     "tsx": "^4.20.6",
     "typescript": "^5.9.3",
+    "vite": "^6.0.1",
+    "vite-tsconfig-paths": "^5.1.2",
     "vitest": "^4.0.6"
   },
   "keywords": [],
@@ -45,7 +47,7 @@
   "scripts": {
     "start": "tsx packages/cli/main.tsx --",
     "build": "tsup --config tsup.config.ts && node scripts/copy-assets.js",
-    "prepublishOnly": "yarn build",
+    "prepack": "yarn build",
     "start:test": "concurrently --raw --kill-others --success first \"yarn start\" \"sleep 10\"",
     "test:story": "yarn start -- --blueprint -p \"Say [Hi]. Don't say anything else.\" --auto-allow --auto-exit",
     "test:glossary": "yarn start -- --glossary -p \"Say [Hi]. Don't say anything else.\" --auto-allow --auto-exit",
@@ -57,7 +59,7 @@
     "test:devhub:add_pending": "concurrently --raw --kill-others --success first \"yarn start -- --workspace $(mktemp -d)/taskagent-tmp --newsession --devhub --auto-allow --auto-exit -p 'Hello, please use send_to_looper tool with command add_pending and task Refactor logging system'\" \"sleep 20\"",
     "test:devhub:stop": "concurrently --raw --kill-others --success first \"yarn start -- --workspace $(mktemp -d)/taskagent-tmp --newsession --devhub --auto-allow --auto-exit -p 'Hello, please use send_to_looper tool with command stop'\" \"sleep 20\"",
     "test:watch": "vitest watch",
-    "publish": "echo \"Use yarn npm publish --access public\"",
+    "publish": "yarn npm publish --access public",
     "test:pack": "node scripts/test-pack.js"
   },
   "_note_postbuild": "Yarn 4 hardened mode skips postbuild hooks, so asset copy runs inline in the build script.",

packages/agents/blueprint/BlueprintLoop.ts

@@ -11,8 +11,8 @@ import yaml from 'js-yaml';
 import { LoopAgent } from '../workflow-agents/LoopAgent.js';
 import type { RunnableAgent, AgentStartContext, AgentStartSinks, AgentToolContext } from '../runtime/types.js';
 import type { AgentRegistry } from '../registry/AgentRegistry.js';
-import type { EventBus } from '@taskagent/core/event-bus';
-import { addLog } from '@taskagent/shared/logger';
+import type { EventBus } from '@core/event-bus';
+import { addLog } from '@shared/logger';
 import { runAgent, emitProgress } from '../runtime/async-task/helpers.js';
 import type { AsyncTaskContext } from '../runtime/async-task/types.js';
 

packages/agents/blueprint/index.ts

@@ -1,5 +1,6 @@
 import type { RunnableAgent } from '../runtime/types.js';
-import type { EventBus } from '@taskagent/core/event-bus';
+
+import type { EventBus } from '@core/event-bus';
 import { BlueprintLoop } from './BlueprintLoop.js';
 
 export async function createAgent(options?: {

packages/agents/coder/CoderAgent.ts

@@ -1,7 +1,7 @@
 import { PromptAgent, type AgentContext, type AgentToolContext, type RunnableAgent, type AgentStartContext, type AgentStartSinks, type ExecutionHandle } from '../runtime/types.js';
 import { buildPromptAgentStart } from '../runtime/runPromptAgentStart.js';
 import type { AgentRegistry } from '../registry/AgentRegistry.js';
-import type { EventBus } from '@taskagent/core/event-bus';
+import type { EventBus } from '@core/event-bus';
 import type { AgentDefinition } from '@anthropic-ai/claude-agent-sdk';
 
 const CODER_AGENT_ID = 'coder';

packages/agents/coder/index.ts

@@ -3,7 +3,7 @@ import { fileURLToPath } from 'url';
 import { loadAgentPipelineConfig } from '../runtime/agentLoader.js';
 import type { RunnableAgent } from '../runtime/types.js';
 import type { TabExecutor } from '../../execution/TabExecutor.js';
-import type { EventBus } from '@taskagent/core/event-bus';
+import type { EventBus } from '@core/event-bus';
 import type { AgentRegistry } from '../registry/AgentRegistry.js';
 import { CoderAgent } from './CoderAgent.js';
 

packages/agents/desktop/index.ts

@@ -15,8 +15,8 @@ import { buildPromptAgentStart } from '../runtime/runPromptAgentStart.js';
 import { loadAgentPipelineConfig } from '../runtime/agentLoader.js';
 import type { tool as createSdkTool } from '@anthropic-ai/claude-agent-sdk';
 import type { AgentContext, AgentStartContext, AgentStartSinks, ExecutionHandle, RunnableAgent } from '../runtime/types.js';
-import type { EventBus } from '@taskagent/core/event-bus';
-import { addLog } from '@taskagent/shared/logger';
+import type { EventBus } from '@core/event-bus';
+import { addLog } from '@shared/logger';
 import type { AgentRegistry } from '../registry/AgentRegistry.js';
 
 const START_AGENT_ID = 'start';

packages/agents/devhub/DevHub.ts

@@ -14,12 +14,12 @@ import { PromptAgent, type AgentContext, type AgentToolContext, type RunnableAge
 import { buildPromptAgentStart } from '../runtime/runPromptAgentStart.js';
 import { loadAgentPipelineConfig } from '../runtime/agentLoader.js';
 import type { AgentDefinition } from '@anthropic-ai/claude-agent-sdk';
-import type { EventBus } from '@taskagent/core/event-bus';
-import type { TaskManager } from '@taskagent/shared/task-manager';
+import type { EventBus } from '@core/event-bus';
+import type { TaskManager } from '@shared/task-manager';
 import type { AgentRegistry } from '../registry/AgentRegistry.js';
 import { CodingLoop } from './coding-loop/index.js';
 import type { LooperCommand } from './coding-loop/command.js';
-import { addLog } from '@taskagent/shared/logger';
+import { addLog } from '@shared/logger';
 
 const DEV_HUB_AGENT_ID = 'devhub';
 const DEV_HUB_DESCRIPTION = 'DevHub Agent - 理解开发需求，协调 Coder 与 Reviewer 循环开发直到代码通过审查';

packages/agents/devhub/coding-loop/SummarizationCallback.ts

@@ -8,7 +8,7 @@
 import type { AgentCallback } from '../../workflow-agents/AgentCallback.js';
 import type { RunnableAgent } from '../../runtime/types.js';
 import { EventCollector } from './event-collector.js';
-import { addLog } from '@taskagent/shared/logger';
+import { addLog } from '@shared/logger';
 
 export class SummarizationCallback implements AgentCallback {
     private eventCollector = new EventCollector();

packages/agents/devhub/coding-loop/event-collector.ts

@@ -7,7 +7,7 @@
  * - Determine when to trigger summary (time-based or count-based)
  */
 
-import { addLog } from '@taskagent/shared/logger';
+import { addLog } from '@shared/logger';
 
 export interface TruncatedEvent {
     type: string;

packages/agents/devhub/coding-loop/index.ts

@@ -17,7 +17,7 @@ import { createJudgeAgent, parseJudgeOutput } from '../judge/index.js';
 import { createSummarizerAgent } from '../summarizer/index.js';
 import { SummarizationCallback } from './SummarizationCallback.js';
 import { SinglePass } from './SinglePass.js';
-import { addLog } from '@taskagent/shared/logger';
+import { addLog } from '@shared/logger';
 import path from 'node:path';
 import { fileURLToPath } from 'node:url';
 

packages/agents/devhub/judge/JudgeAgent.ts

@@ -12,7 +12,7 @@ import { fileURLToPath } from 'url';
 import { PromptAgent, type AgentContext, type AgentToolContext, type RunnableAgent, type AgentStartContext, type AgentStartSinks, type ExecutionHandle } from '../../runtime/types.js';
 import { buildPromptAgentStart } from '../../runtime/runPromptAgentStart.js';
 import { JudgeDecisionSchema, type JudgeDecision } from './schema.js';
-import { addLog } from '@taskagent/shared/logger';
+import { addLog } from '@shared/logger';
 import fs from 'fs/promises';
 
 const JUDGE_AGENT_ID = 'judge';