Implements 123 tasks across 9 phases (T-001..T-809) totaling 146 source files. Monorepo (P0): - 7-package Bun + Turborepo + TypeScript monorepo - dependency-cruiser enforcing 7 forbidden edges + 5 deep-import rules Contracts (P0): - 16 type files (ids/error/event/runtime/ipc/task/worker-result/tool/artifact/evidence/project/provider/permission/ui/capability/platform) Storage & Events (P1): - DatabaseManager + MigrationRunner (19 tables, 22 indexes, 5 schema_meta seeds) - 16 repositories (Repository<T,I,U> pattern, INV-1 status columns via EventStore.project only) - EventSchemaRegistry (54 durable + 7 ephemeral), EventStore, EventBus, EventIngestor - Project/Session/Artifact/Evidence stores + 8-step Recovery Tools & Permission (P2): - PathClassifier (8 categories), CommandRiskAnalyzer (10 categories), SecretRedactor - PermissionEngine 6-layer evaluation (capability→profile→task_scope→risk→credential→user_prompt) - ToolRegistry with 20+ tools across fs/shell/git/project/artifact/context/permission/doctor - CapabilityManifestValidator + CapabilityRegistry LLM & Context (P3): - ModelConfigLoader, CapabilityMatrix, AnthropicCanonicalConverter - AnthropicAdapter + OpenAICompatibleAdapter - ProviderManager facade - PromptLayerLoader (L0/L1/L3/L5), CompactionPolicy, ContextAssembler Worker IPC & Scheduler (P4): - WorkerProtocol (NDJSON), WorkerProcess (exit codes 0-5), WorkerManager (spawn/handshake) - WorkerRuntime (INV-3: IPC only, no direct fs/shell/SQLite) - 5 worker roles (Executor/Reviewer/Debugger/Compactor/ExperienceMiner) - TaskGraph, WavePlanner, RetryPlanner, AgentMonitor, WorkspaceManager - Scheduler (state machine), 8-step Recovery C++ Toolchain (P5): - DiagnosticParser, CppProjectDetector, CMakeConfigurator, CppBuilder - CppTestRunner, CppcheckRunner, ClangdClient - CppToolRegistrar + capability manifest Projection & TUI (P6): - ProjectionStore (hydrate/apply/snapshot/subscribe) - TuiApp + 8 components (Session/Task/Agent/Tool/Diff/Evidence/Permission/Blocker/Hud) - ProjectionClient in-process ref Agents & Knowledge (P7): - MainAgent, ArchitectureDesigner - DebugKnowledgeStore + LearnedMemoryStore (single-writer, outbox model) - Role integration wiring CLI & Doctor & Release (P8): - Logger + DeveloperLogEncryptor (AES-256-GCM) - DoctorService (self_bootstrap first) - RuntimeApp + ServiceRegistry - 11 CLI commands: run/init/doctor/provider/resume/compact/history/session/restore/e2e/release - CliEntrypoint + air<TODO> Audit (in AirPlan/docs/): - Deepseek开发阶段审计.md (97 findings) - Opus开发阶段审计.md (140+ findings, 18 P0 blockers) - MiniMaxM3开发阶段审计.md (18 P0 blockers, focuses on executability) - AirPlan/TODO.md (technical debt + 42 TODOs by phase) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
339 lines
11 KiB
TypeScript
Executable File
339 lines
11 KiB
TypeScript
Executable File
/**
|
|
* ToolRegistry - tool lookup, validation, permission, execution
|
|
*
|
|
* Implements contracts §12; DD §9.1 + §9.3 branching table.
|
|
* INV-3: all tool execution must go through this registry.
|
|
*
|
|
* @module packages/runtime/src/tools/ToolRegistry
|
|
*/
|
|
|
|
import type { ToolDefinition, ToolCall, ToolResultEnvelope, ISOTimeString } from '@aircoding/contracts'
|
|
|
|
import { PermissionEngine, createPermissionEngine, type PermissionContext, type PermissionDecision, type PermissionAction } from '../security/PermissionEngine.js'
|
|
import type { AgentType } from '@aircoding/contracts'
|
|
|
|
export interface ToolExecutor {
|
|
(call: ToolCall, context: ToolExecutionContext): Promise<ToolResultEnvelope>
|
|
}
|
|
|
|
export interface ToolExecutionContext {
|
|
session_id: string
|
|
project_id: string
|
|
project_root: string
|
|
agent_id: string
|
|
agent_type: AgentType
|
|
}
|
|
|
|
export interface ToolCallContext {
|
|
tool_definition: ToolDefinition
|
|
executor: ToolExecutor
|
|
permission_context: PermissionContext
|
|
}
|
|
|
|
/**
|
|
* Branching behavior per DD §9.3
|
|
*/
|
|
const ACTION_BRANCHES: Record<PermissionAction, (decision: PermissionDecision, call: ToolCall, ctx: ToolExecutionContext) => Promise<ToolResultEnvelope>> = {
|
|
allow: async (_decision, call, ctx) => {
|
|
// Execute directly
|
|
const definition = global_tool_registry?.get(call.name)
|
|
if (!definition) {
|
|
return create_error_result(call.id, 'tool_not_found', 'Tool not found')
|
|
}
|
|
const executor = global_tool_registry?.executors.get(call.name)
|
|
if (!executor) {
|
|
return create_error_result(call.id, 'executor_not_found', 'Executor not registered')
|
|
}
|
|
return executor(call, ctx)
|
|
},
|
|
|
|
deny: async (decision) => {
|
|
return create_error_result('', 'permission_denied', decision.reason)
|
|
},
|
|
|
|
prompt: async (_decision, _call, _ctx) => {
|
|
// TODO: Integrate with UI for user prompt
|
|
// For now, deny with prompt message
|
|
return create_error_result('', 'user_prompt_required', 'User confirmation required')
|
|
},
|
|
|
|
read_only: async (_decision, call, ctx) => {
|
|
// Downgrade write operations to read-only
|
|
const modified_call = this.downgrade_to_readonly(call)
|
|
const definition = global_tool_registry?.get(call.name)
|
|
const executor = global_tool_registry?.executors.get(call.name)
|
|
if (!executor) {
|
|
return create_error_result(call.id, 'executor_not_found', 'Executor not registered')
|
|
}
|
|
return executor(modified_call as ToolCall, ctx)
|
|
},
|
|
|
|
sandbox: async (decision, call, ctx) => {
|
|
// Execute in sandboxed mode with restricted environment
|
|
const sandboxed_call = {
|
|
...call,
|
|
arguments: this.apply_sandbox_restrictions(call.arguments, decision.flags)
|
|
}
|
|
const executor = global_tool_registry?.executors.get(call.name)
|
|
if (!executor) {
|
|
return create_error_result(call.id, 'executor_not_found', 'Executor not registered')
|
|
}
|
|
return executor(sandboxed_call, ctx)
|
|
},
|
|
|
|
audit_log: async (_decision, call, ctx) => {
|
|
// Execute and log for audit
|
|
const definition = global_tool_registry?.get(call.name)
|
|
const executor = global_tool_registry?.executors.get(call.name)
|
|
if (!executor) {
|
|
return create_error_result(call.id, 'executor_not_found', 'Executor not registered')
|
|
}
|
|
const result = await executor(call, ctx)
|
|
// Add audit flag to result
|
|
return {
|
|
...result,
|
|
metadata: { ...result.metadata, audit_logged: true }
|
|
}
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Global tool registry (singleton)
|
|
*/
|
|
let global_tool_registry: ToolRegistry | undefined
|
|
|
|
export class ToolRegistry {
|
|
private tools: Map<string, ToolDefinition> = new Map()
|
|
private executors: Map<string, ToolExecutor> = new Map()
|
|
private permission_engine: PermissionEngine
|
|
private project_root: string
|
|
|
|
constructor(project_root: string) {
|
|
this.project_root = project_root
|
|
this.permission_engine = createPermissionEngine(project_root)
|
|
global_tool_registry = this
|
|
}
|
|
|
|
/**
|
|
* Register a tool with its definition and executor.
|
|
*/
|
|
register(name: string, definition: ToolDefinition, executor: ToolExecutor): void {
|
|
this.tools.set(name, definition)
|
|
this.executors.set(name, executor)
|
|
}
|
|
|
|
/**
|
|
* Unregister a tool.
|
|
*/
|
|
unregister(name: string): void {
|
|
this.tools.delete(name)
|
|
this.executors.delete(name)
|
|
}
|
|
|
|
/**
|
|
* Get tool definition by name.
|
|
*/
|
|
get(name: string): ToolDefinition | undefined {
|
|
return this.tools.get(name)
|
|
}
|
|
|
|
/**
|
|
* List all registered tools.
|
|
*/
|
|
list(): ToolDefinition[] {
|
|
return Array.from(this.tools.values())
|
|
}
|
|
|
|
/**
|
|
* Call a tool with permission evaluation and branching.
|
|
* Implements DD §9.1 algorithm.
|
|
*/
|
|
async call(call: ToolCall, context: ToolExecutionContext): Promise<ToolResultEnvelope> {
|
|
// Step 1: Lookup tool definition
|
|
const definition = this.tools.get(call.name)
|
|
if (!definition) {
|
|
return create_error_result(call.id, 'tool_not_found', `Tool ${call.name} not found`)
|
|
}
|
|
|
|
// Step 2: Validate input schema
|
|
const validation = this.validate_input(call, definition)
|
|
if (!validation.valid) {
|
|
return create_error_result(call.id, 'invalid_input', validation.error || 'Invalid input')
|
|
}
|
|
|
|
// Step 3: Build permission context
|
|
const permission_context = this.build_permission_context(call, context)
|
|
|
|
// Step 4: Evaluate permissions (layered per DD §9.2)
|
|
const decision = await this.permission_engine.evaluate(call, permission_context, definition)
|
|
|
|
// Step 5: Branch on permission action (DD §9.3)
|
|
const branch = ACTION_BRANCHES[decision.action]
|
|
if (!branch) {
|
|
return create_error_result(call.id, 'invalid_decision', 'Invalid permission decision')
|
|
}
|
|
|
|
// Step 6: Execute branch
|
|
try {
|
|
const result = await branch(decision, call, context)
|
|
|
|
// Step 7: Record decision (if enabled)
|
|
await this.permission_engine.record(decision)
|
|
|
|
return result
|
|
} catch (error) {
|
|
return create_error_result(call.id, 'execution_error', error instanceof Error ? error.message : String(error))
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Streaming call - returns chunks for real-time output.
|
|
* Ends with exactly one final ToolResultEnvelope.
|
|
*/
|
|
async *call_streaming(call: ToolCall, context: ToolExecutionContext): AsyncGenerator<ToolResultEnvelope> {
|
|
const definition = this.tools.get(call.name)
|
|
if (!definition?.streaming) {
|
|
// Non-streaming tool, call normally and yield single result
|
|
const result = await this.call(call, context)
|
|
yield result
|
|
return
|
|
}
|
|
|
|
// For streaming tools, we need to get the executor
|
|
const executor = this.executors.get(call.name)
|
|
if (!executor) {
|
|
yield create_error_result(call.id, 'executor_not_found', 'Executor not registered')
|
|
return
|
|
}
|
|
|
|
// Permission check first (same as call)
|
|
const permission_context = this.build_permission_context(call, context)
|
|
const decision = await this.permission_engine.evaluate(call, permission_context, definition)
|
|
|
|
if (decision.action !== 'allow') {
|
|
yield create_error_result(call.id, 'permission_denied', decision.reason)
|
|
return
|
|
}
|
|
|
|
// Execute with streaming support
|
|
// The executor yields intermediate results, final result comes at end
|
|
let final_result: ToolResultEnvelope | undefined
|
|
|
|
for await (const chunk of this.execute_streaming(call, context, executor)) {
|
|
if (chunk.type === 'final') {
|
|
final_result = chunk
|
|
} else {
|
|
yield chunk
|
|
}
|
|
}
|
|
|
|
// Yield final result exactly once
|
|
if (final_result) {
|
|
yield final_result
|
|
} else {
|
|
yield create_error_result(call.id, 'no_final_result', 'Streaming tool did not produce final result')
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Validate tool input against schema.
|
|
*/
|
|
private validate_input(call: ToolCall, definition: ToolDefinition): { valid: boolean; error?: string } {
|
|
// Basic validation - in production, use JSON Schema validation
|
|
if (!call.arguments || typeof call.arguments !== 'object') {
|
|
return { valid: false, error: 'Arguments must be an object' }
|
|
}
|
|
|
|
// Check required fields if specified in definition
|
|
// This is a simplified check - full implementation would use JSON Schema
|
|
return { valid: true }
|
|
}
|
|
|
|
/**
|
|
* Build permission context from tool call and execution context.
|
|
*/
|
|
private build_permission_context(call: ToolCall, context: ToolExecutionContext): PermissionContext {
|
|
return {
|
|
session_id: context.session_id,
|
|
project_id: context.project_id,
|
|
project_root: context.project_root,
|
|
agent_type: context.agent_type,
|
|
agent_id: context.agent_id,
|
|
task_scope: undefined, // Would be loaded from task context
|
|
permission_profile: undefined // Would be loaded from agent config
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Downgrade write operations to read-only.
|
|
*/
|
|
private downgrade_to_readonly(call: ToolCall): ToolCall {
|
|
// Modify arguments to make operation read-only
|
|
const modified = { ...call.arguments }
|
|
|
|
// For filesystem operations, remove write-related flags
|
|
if ('mode' in modified && typeof modified.mode === 'string') {
|
|
if (modified.mode.includes('w') || modified.mode.includes('a')) {
|
|
modified.mode = modified.mode.replace(/[wa]/g, 'r')
|
|
}
|
|
}
|
|
|
|
// Remove force flags
|
|
delete modified.force
|
|
delete modified.overwrite
|
|
|
|
return { ...call, arguments: modified }
|
|
}
|
|
|
|
/**
|
|
* Apply sandbox restrictions to arguments.
|
|
*/
|
|
private apply_sandbox_restrictions(args: Record<string, unknown>, flags: string[]): Record<string, unknown> {
|
|
const restricted = { ...args }
|
|
|
|
// Add sandbox restrictions based on flags
|
|
if (flags.includes('no_network')) {
|
|
delete restricted.url
|
|
delete restricted.endpoint
|
|
}
|
|
|
|
if (flags.includes('read_only')) {
|
|
// Already handled in read_only branch
|
|
}
|
|
|
|
// Add sandbox metadata
|
|
return restricted
|
|
}
|
|
|
|
/**
|
|
* Execute streaming tool.
|
|
*/
|
|
private async *execute_streaming(
|
|
call: ToolCall,
|
|
context: ToolExecutionContext,
|
|
executor: ToolExecutor
|
|
): AsyncGenerator<ToolResultEnvelope> {
|
|
// This is a placeholder - actual implementation would depend on the tool
|
|
// For now, just execute normally
|
|
const result = await executor(call, context)
|
|
yield result
|
|
}
|
|
}
|
|
|
|
export function createToolRegistry(project_root: string): ToolRegistry {
|
|
return new ToolRegistry(project_root)
|
|
}
|
|
|
|
// ============================================================================
|
|
// Result helpers
|
|
// ============================================================================
|
|
|
|
function create_error_result(call_id: string, error_type: string, message: string): ToolResultEnvelope {
|
|
return {
|
|
call_id,
|
|
tool_name: '',
|
|
type: 'error',
|
|
content: { error_type, message },
|
|
metadata: { timestamp: new Date().toISOString() as ISOTimeString }
|
|
}
|
|
} |