thumbgate 1.29.1 → 1.30.0

package/scripts/tool-registry.js

@@ -60,8 +60,93 @@ const GOAL_CONTRACT_SCHEMA = {
   },
 };
 
+const TASK_OUTCOME_INPUT_SCHEMA = {
+  type: 'object',
+  additionalProperties: false,
+  required: ['taskId', 'goal', 'status', 'verification', 'toolCalls', 'policy', 'efficiency'],
+  properties: {
+    taskId: { type: 'string', minLength: 1 },
+    taskType: { type: 'string' },
+    goal: { type: 'string', minLength: 1 },
+    expectedOutcome: { type: 'string' },
+    status: { type: 'string', enum: ['completed', 'failed', 'partial', 'escalated'] },
+    verification: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['performed', 'passed', 'evidence'],
+      properties: {
+        performed: { type: 'boolean' },
+        passed: { type: 'boolean' },
+        verifier: { type: 'string' },
+        method: { type: 'string' },
+        evidence: { type: 'array', items: { type: 'string' } },
+        unsupportedClaims: { type: 'integer', minimum: 0 },
+      },
+    },
+    toolCalls: {
+      type: 'array',
+      items: {
+        type: 'object',
+        additionalProperties: false,
+        required: ['name', 'contractValid', 'allowed', 'succeeded', 'attempts'],
+        properties: {
+          name: { type: 'string' },
+          contractValid: { type: 'boolean' },
+          allowed: { type: 'boolean' },
+          succeeded: { type: 'boolean' },
+          attempts: { type: 'integer', minimum: 1 },
+          latencyMs: { type: 'number', minimum: 0 },
+          costUsd: { type: 'number', minimum: 0 },
+          sideEffect: { type: 'boolean' },
+          idempotencyKey: { type: 'string' },
+          duplicateSideEffect: { type: 'boolean' },
+        },
+      },
+    },
+    policy: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['violations', 'unsafeEscapes', 'falseBlocks'],
+      properties: {
+        violations: { type: 'integer', minimum: 0 },
+        unsafeEscapes: { type: 'integer', minimum: 0 },
+        falseBlocks: { type: 'integer', minimum: 0 },
+      },
+    },
+    failure: { type: 'object', additionalProperties: true },
+    escalation: { type: 'object', additionalProperties: true },
+    efficiency: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['latencyMs', 'costUsd'],
+      properties: {
+        latencyMs: { type: 'number', minimum: 0 },
+        costUsd: { type: 'number', minimum: 0 },
+        firstAttempt: { type: 'boolean' },
+      },
+    },
+    businessOutcome: { type: 'object', additionalProperties: true },
+    traceId: { type: 'string' },
+    idempotencyKey: { type: 'string' },
+    versions: { type: 'object', additionalProperties: true },
+    metadata: { type: 'object', additionalProperties: true },
+  },
+};
+
+const MEMORY_SCOPE_SCHEMA = {
+  type: 'object',
+  additionalProperties: false,
+  required: ['entityId', 'projectId', 'processId', 'sessionId'],
+  properties: {
+    entityId: { type: 'string', minLength: 1 },
+    projectId: { type: 'string', minLength: 1 },
+    processId: { type: 'string', minLength: 1 },
+    sessionId: { type: 'string', minLength: 1 },
+  },
+};
+
 const TOOLS = [
-  readOnlyTool({
+  destructiveTool({
     name: 'capture_feedback',
     description: 'Capture an up/down signal plus one line of why. Vague feedback is logged, then returned with a clarification prompt instead of memory promotion.',
     inputSchema: {
@@ -145,6 +230,9 @@ const TOOLS = [
         limit: { type: 'number', description: 'Maximum results to return (default 10)' },
         category: { type: 'string', enum: ['error', 'learning', 'preference'] },
         tags: { type: 'array', items: { type: 'string' }, description: 'Require all tags to be present on a lesson' },
+        scope: MEMORY_SCOPE_SCHEMA,
+        requireScope: { type: 'boolean', description: 'Fail closed unless a complete four-field scope is supplied.' },
+        includeShared: { type: 'boolean', description: 'Include explicitly shared memories with scoped results. Defaults true.' },
       },
     },
   }),
@@ -157,6 +245,9 @@ const TOOLS = [
         toolName: { type: 'string', description: 'The tool being called (e.g., Bash, Edit, Read)' },
         actionContext: { type: 'string', description: 'Description of what the tool call is doing' },
         maxResults: { type: 'number', description: 'Max lessons to return (default 5)' },
+        scope: MEMORY_SCOPE_SCHEMA,
+        requireScope: { type: 'boolean', description: 'Fail closed unless a complete four-field scope is supplied.' },
+        includeShared: { type: 'boolean', description: 'Include explicitly shared memories with scoped results. Defaults true.' },
       },
       required: ['toolName'],
     },
@@ -831,6 +922,10 @@ const TOOLS = [
           items: { type: 'string' },
           description: 'Optional protected-file globs that require explicit approval before editing or publishing',
         },
+        ttlMs: {
+          type: 'number',
+          description: 'Optional lease length in milliseconds. With it the scope becomes time-bounded authority (e.g. 90000 for "write under ./src for 90 seconds") and FAILS CLOSED on expiry: a lapsed lease authorises nothing until renewed. Omit for a permanent scope. Clamped to 60s..24h.',
+        },
         workflowContract: {
           type: 'object',
           description: 'Optional deterministic workflow run contract. Supports workflowId, allowedBranches, blockedActions, requiredEvidence, and completionGate.',
@@ -969,6 +1064,102 @@ const TOOLS = [
       },
     },
   }),
+  destructiveTool({
+    name: 'record_task_outcome',
+    title: 'Record Verified Task Outcome',
+    description: 'Record an idempotent task-level outcome with verification evidence, tool correctness, policy behavior, latency, cost, and business KPI movement. A completed response without evidence is recorded as not working.',
+    inputSchema: TASK_OUTCOME_INPUT_SCHEMA,
+    outputSchema: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['recorded', 'duplicate', 'receipt'],
+      properties: {
+        recorded: { type: 'boolean' },
+        duplicate: { type: 'boolean' },
+        receipt: { type: 'object', additionalProperties: true },
+      },
+    },
+  }),
+  readOnlyTool({
+    name: 'get_task_outcomes',
+    title: 'Get Task Outcomes',
+    description: 'Read a task outcome by taskId or the most recent evidence-backed outcome receipts.',
+    inputSchema: {
+      type: 'object',
+      additionalProperties: false,
+      properties: {
+        taskId: { type: 'string' },
+        limit: { type: 'integer', minimum: 1, maximum: 100 },
+      },
+    },
+  }),
+  readOnlyTool({
+    name: 'get_agent_outcome_metrics',
+    title: 'Get Agent Outcome Metrics',
+    description: 'Compute transparent task, tool, safety, escalation, latency, cost, and business metrics from recorded task outcomes. Empty data returns insufficient_evidence.',
+    inputSchema: {
+      type: 'object',
+      additionalProperties: false,
+      properties: {},
+    },
+    outputSchema: {
+      type: 'object',
+      additionalProperties: true,
+      required: ['generatedAt', 'sampleSize', 'evidenceStatus', 'task', 'tools', 'safety', 'escalation', 'efficiency', 'businessOutcomes'],
+      properties: {
+        generatedAt: { type: 'string', format: 'date-time' },
+        sampleSize: { type: 'integer', minimum: 0 },
+        evidenceStatus: { type: 'string', enum: ['measured', 'insufficient_evidence'] },
+        task: { type: 'object', additionalProperties: true },
+        tools: { type: 'object', additionalProperties: true },
+        safety: { type: 'object', additionalProperties: true },
+        escalation: { type: 'object', additionalProperties: true },
+        efficiency: { type: 'object', additionalProperties: true },
+        businessOutcomes: { type: 'array' },
+      },
+    },
+  }),
+  destructiveTool({
+    name: 'request_human_escalation',
+    title: 'Request Human Escalation',
+    description: 'Create an idempotent, expiring human-review request with requester identity and evidence. Agents cannot approve their own requests.',
+    inputSchema: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['taskId', 'reason', 'requester', 'evidence'],
+      properties: {
+        taskId: { type: 'string', minLength: 1 },
+        reason: { type: 'string', minLength: 1 },
+        severity: { type: 'string', enum: ['low', 'medium', 'high', 'critical'] },
+        requester: {
+          type: 'object',
+          additionalProperties: false,
+          required: ['id', 'kind'],
+          properties: {
+            id: { type: 'string', minLength: 1 },
+            kind: { type: 'string', enum: ['agent', 'service', 'human'] },
+            displayName: { type: 'string' },
+          },
+        },
+        evidence: { type: 'array', minItems: 1, items: { type: 'string', minLength: 1 } },
+        ttlMs: { type: 'number', minimum: 1 },
+        idempotencyKey: { type: 'string' },
+      },
+    },
+  }),
+  readOnlyTool({
+    name: 'list_human_escalations',
+    title: 'List Human Escalations',
+    description: 'List auditable human-escalation state. Approval decisions are deliberately absent from the agent tool surface.',
+    inputSchema: {
+      type: 'object',
+      additionalProperties: false,
+      properties: {
+        status: { type: 'string', enum: ['pending', 'approved', 'rejected', 'cancelled', 'expired'] },
+        limit: { type: 'integer', minimum: 1, maximum: 100 },
+      },
+    },
+  }),
   readOnlyTool({
     name: 'verify_claim',
     description: 'Check whether a claim has enough tracked evidence before the agent asserts it.',