Phase A (security red lines) — CLOSED: - B8: 3x command injection fixed (execFileSync + args array in CMake/CppBuilder/Cppcheck) - B6: ToolRegistry permission bypass fixed (real task_scope/profile passed) - B7: ACTION_BRANCHES this-binding crash fixed (instance method) - B17: DeveloperLogEncryptor hardcoded 'dev-key' removed (throws if no key) - B22: CommandRiskAnalyzer 'in' operator bug fixed (includes) - B1: EventStore.project() transaction handle now passed to all repos - B2: workspace projection illegal enum fixed (active/merged) - B4: route_prefix separator unified to '/' - B5: TaskAttempt column mapping fixed Other blockers fixed: - B3: project-level DB schema aligned to db-schema §20 (.air/local, learned_memories) - B9: cpp.* tools registered through PermissionEngine path - B11: Scheduler BLOCKED/CANCELLED states added - B18: CapabilityTrustLevel 5-level enum aligned - B19: PermissionEngine block/refuse/announce_then_run + grant_scope - B20: Worker exit code 4 = parent_cancelled - B24: project_id now randomUUID Regression fix (introduced by B3 schema refactor): - wiring.ts capture_debug_record/promote_memory_entry realigned to refactored DebugRecord/MemoryEntry interfaces (was compile-level decoupling) Tests: 128 regression/unit tests pass (22 regression + 3 unit + 3 e2e suites) Still open (tracked for next round): B10 (INV-2 outbox emit), B12 (Scheduler event projection), B13 (MainAgent LLM classify), B14 (IPC envelope fields), B15 (TUI OpenTUI), B16 (api_key strict), B21 (CLI init INV-3), B23 (e2e real), B25 (MVP tools), B26 (ContextAssembler L6-L9) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
322 lines
9.8 KiB
TypeScript
Executable File
322 lines
9.8 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
|
|
task_scope?: PermissionContext['task_scope']
|
|
permission_profile?: PermissionContext['permission_profile']
|
|
}
|
|
|
|
export interface ToolCallContext {
|
|
tool_definition: ToolDefinition
|
|
executor: ToolExecutor
|
|
permission_context: PermissionContext
|
|
}
|
|
|
|
/**
|
|
* 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)
|
|
// Step 6: Execute branch
|
|
try {
|
|
const result = await this.execute_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: context.task_scope,
|
|
permission_profile: context.permission_profile,
|
|
}
|
|
}
|
|
|
|
/**
|
|
* 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 branching behavior per DD §9.3.
|
|
* Replaces the module-level ACTION_BRANCHES to fix `this` binding.
|
|
*/
|
|
private async execute_branch(
|
|
decision: PermissionDecision,
|
|
call: ToolCall,
|
|
ctx: ToolExecutionContext,
|
|
): Promise<ToolResultEnvelope> {
|
|
switch (decision.action) {
|
|
case 'allow': {
|
|
const executor = this.executors.get(call.name)
|
|
if (!executor) {
|
|
return create_error_result(call.id, 'executor_not_found', 'Executor not registered')
|
|
}
|
|
return executor(call, ctx)
|
|
}
|
|
|
|
case 'announce_then_run': {
|
|
// Emit visible notice, then execute unless interrupted
|
|
const executor = this.executors.get(call.name)
|
|
if (!executor) {
|
|
return create_error_result(call.id, 'executor_not_found', 'Executor not registered')
|
|
}
|
|
const result = await executor(call, ctx)
|
|
return {
|
|
...result,
|
|
metadata: { ...result.metadata, announced: true },
|
|
}
|
|
}
|
|
|
|
case 'ask_user':
|
|
// Suspend; emit permission.prompt.requested
|
|
return create_error_result('', 'user_prompt_required', 'User confirmation required')
|
|
|
|
case 'deny':
|
|
return create_error_result('', 'permission_denied', decision.reason)
|
|
|
|
case 'block': {
|
|
// Return blocked outcome → task.blocked upstream
|
|
return create_error_result(call.id, 'blocked', `Action blocked: ${decision.reason}`)
|
|
}
|
|
|
|
case 'refuse': {
|
|
// Return policy error; no execution
|
|
return create_error_result(call.id, 'policy_error', `Refused: ${decision.reason}`)
|
|
}
|
|
|
|
default:
|
|
return create_error_result(call.id, 'invalid_decision', `Unknown action: ${decision.action}`)
|
|
}
|
|
}
|
|
|
|
/**
|
|
* 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 }
|
|
}
|
|
} |