@zibby/core 0.3.1 → 0.3.3

package/dist/templates/browser-test-automation/nodes/execute-live.mjs

@@ -74,6 +74,23 @@ DO NOT:
 - Spend more than 2 minutes on any single page
 - Try to click elements that aren't immediately visible
 
+📋 BEFORE EXECUTING (MEMORY-AWARE START):
+If the "Domain Knowledge" section above contains a "### Known Pages on This Site"
+block with "expected fingerprint: zibby-..." entries for the URL you're about
+to land on, do this ONCE on first navigation:
+
+  1. Navigate to the page.
+  2. Call browser_snapshot just once to read the live DOM's stableIds.
+  3. Count how many of the expected fingerprint stableIds are present.
+  4. If ≥ 85% are present → MATCH: trust the cached selectors directly,
+     no further exploration on this page is needed.
+  5. If < 85% are present → MISMATCH: the page has drifted, ignore the
+     cached selectors and rediscover from the live snapshot.
+
+This costs ONE snapshot per page in exchange for skipping the rest. Don't
+skip this step on pages that have an expected fingerprint — it's the
+mechanism that makes repeat runs cheap.
+
 EXECUTION SEQUENCE (MANDATORY - FOLLOW STRICTLY):
 1. Execute the test steps efficiently (navigate, fill, click)
    - Max 10-15 actions total
@@ -97,6 +114,15 @@ IMPORTANT for 'actions' array (STRICT 1:1 MAPPING):
 - If you call browser_type 3 times for 3 fields, you MUST have 3 actions in the array.
 - Include actual values/URLs in descriptions.
 - Keep descriptions SHORT (5-10 words max).
+- **'committed' field (REQUIRED on actions you DELIBERATELY chose):**
+  set committed: true when this action was your CHOSEN attempt for an
+  intent (the click/fill you actually wanted, regardless of outcome).
+  set committed: false (or omit) for exploratory probes you tried while
+  searching for the right element. Only committed actions feed the
+  negative cache, so probes don't pollute future runs' Avoid lists.
+  Example: tried selector A (probe, failed) → tried selector B
+  (deliberate, failed) → tried selector C (deliberate, succeeded). Mark
+  B and C as committed: true; A as committed: false.
 
 IMPORTANT for 'assertions' array (USE THE CHECKLIST ABOVE):
 - Your assertions array MUST match the ASSERTION CHECKLIST exactly - one entry per item
@@ -164,6 +190,12 @@ IMPORTANT for 'evidenceScreenshots' (array) - OPTIONAL:
         .describe('Human-readable description of the action'),
       reasoning: z.string().nullish()
         .describe('Why this action was performed'),
+      committed: z.boolean().nullish()
+        .describe('true when this was a deliberate chosen attempt for an intent (feeds negative cache on failure); false/omit for exploratory probes'),
+      status: z.enum(['success', 'failed']).nullish()
+        .describe('Outcome of the action — set "failed" if the tool call errored or the post-condition was not met'),
+      error: z.string().nullish()
+        .describe('Error message when status=failed'),
       selectors: z.object({
         role: z.object({
           role: z.string().describe('ARIA role (e.g. button, link, textbox, generic)'),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@zibby/core",
-  "version": "0.3.1",
+  "version": "0.3.3",
   "description": "Core test automation engine with multi-agent and multi-MCP support",
   "type": "module",
   "main": "dist/index.js",

package/templates/browser-test-automation/graph.mjs

@@ -43,12 +43,12 @@ export class BrowserTestAutomationAgent extends WorkflowAgent {
       result.state?.cwd || cwd,
     );
 
-    // Memory end-run hook (if @zibby/memory is installed)
+    // Memory end-run hook (if @zibby/ui-memory is installed)
     try {
-      const { memoryEndRun, memorySyncPush } = await import('@zibby/memory');
+      const { memoryEndRun, memorySyncPush } = await import('@zibby/ui-memory');
       const sessionId = result.state.sessionPath?.split('/').pop();
       memoryEndRun(cwd, { sessionId, passed: result.success !== false });
       memorySyncPush(cwd);
-    } catch { /* @zibby/memory not available */ }
+    } catch { /* @zibby/ui-memory not available */ }
   }
 }

package/templates/browser-test-automation/nodes/execute-live.mjs

@@ -74,6 +74,23 @@ DO NOT:
 - Spend more than 2 minutes on any single page
 - Try to click elements that aren't immediately visible
 
+📋 BEFORE EXECUTING (MEMORY-AWARE START):
+If the "Domain Knowledge" section above contains a "### Known Pages on This Site"
+block with "expected fingerprint: zibby-..." entries for the URL you're about
+to land on, do this ONCE on first navigation:
+
+  1. Navigate to the page.
+  2. Call browser_snapshot just once to read the live DOM's stableIds.
+  3. Count how many of the expected fingerprint stableIds are present.
+  4. If ≥ 85% are present → MATCH: trust the cached selectors directly,
+     no further exploration on this page is needed.
+  5. If < 85% are present → MISMATCH: the page has drifted, ignore the
+     cached selectors and rediscover from the live snapshot.
+
+This costs ONE snapshot per page in exchange for skipping the rest. Don't
+skip this step on pages that have an expected fingerprint — it's the
+mechanism that makes repeat runs cheap.
+
 EXECUTION SEQUENCE (MANDATORY - FOLLOW STRICTLY):
 1. Execute the test steps efficiently (navigate, fill, click)
    - Max 10-15 actions total
@@ -97,6 +114,15 @@ IMPORTANT for 'actions' array (STRICT 1:1 MAPPING):
 - If you call browser_type 3 times for 3 fields, you MUST have 3 actions in the array.
 - Include actual values/URLs in descriptions.
 - Keep descriptions SHORT (5-10 words max).
+- **'committed' field (REQUIRED on actions you DELIBERATELY chose):**
+  set committed: true when this action was your CHOSEN attempt for an
+  intent (the click/fill you actually wanted, regardless of outcome).
+  set committed: false (or omit) for exploratory probes you tried while
+  searching for the right element. Only committed actions feed the
+  negative cache, so probes don't pollute future runs' Avoid lists.
+  Example: tried selector A (probe, failed) → tried selector B
+  (deliberate, failed) → tried selector C (deliberate, succeeded). Mark
+  B and C as committed: true; A as committed: false.
 
 IMPORTANT for 'assertions' array (USE THE CHECKLIST ABOVE):
 - Your assertions array MUST match the ASSERTION CHECKLIST exactly - one entry per item
@@ -164,6 +190,12 @@ IMPORTANT for 'evidenceScreenshots' (array) - OPTIONAL:
         .describe('Human-readable description of the action'),
       reasoning: z.string().nullish()
         .describe('Why this action was performed'),
+      committed: z.boolean().nullish()
+        .describe('true when this was a deliberate chosen attempt for an intent (feeds negative cache on failure); false/omit for exploratory probes'),
+      status: z.enum(['success', 'failed']).nullish()
+        .describe('Outcome of the action — set "failed" if the tool call errored or the post-condition was not met'),
+      error: z.string().nullish()
+        .describe('Error message when status=failed'),
       selectors: z.object({
         role: z.object({
           role: z.string().describe('ARIA role (e.g. button, link, textbox, generic)'),

package/dist/templates/code-implementation/graph.js

@@ -1,35 +0,0 @@
-/**
- * Implementation Graph - Apply code changes AND create PR
- * 
- * Flow:
- * 1. setup - Clone repos and init git baseline
- * 2. implement_code - Apply changes, commit, push to branch
- * 3. generate_tests - Generate test cases
- * 4. create_pr - Create GitHub Pull Request
- * 5. finalize - Upload results to API
- */
-
-import { setupNode } from '../code-analysis/nodes/setup-node.js';
-import { implementCodeNode } from '../code-analysis/nodes/generate-code-node.js';
-import { generateTestCasesNode } from '../code-analysis/nodes/generate-test-cases-node.js';
-import { createPRNode } from '../code-analysis/nodes/create-pr-node.js';
-import { finalizeNode } from '../code-analysis/nodes/finalize-node.js';
-import { implementationStateSchema } from './state.js';
-
-export function buildImplementationGraph(graph) {
-  graph.setStateSchema(implementationStateSchema);
-  
-  graph
-    .addNode('setup', setupNode)                   // ← Reused from analysis!
-    .addNode('implement_code', implementCodeNode)  // ← Real changes + push
-    .addNode('generate_test_cases', generateTestCasesNode)  // ← Generate human-readable test specs
-    .addNode('create_pr', createPRNode)            // ← Create GitHub PR
-    .addNode('finalize', finalizeNode)             // ← Reused from analysis!
-    .setEntryPoint('setup')
-    .addEdge('setup', 'implement_code')
-    .addEdge('implement_code', 'generate_test_cases')
-    .addEdge('generate_test_cases', 'create_pr')
-    .addEdge('create_pr', 'finalize');
-  
-  return graph;
-}

package/dist/templates/code-implementation/index.js

@@ -1,7 +0,0 @@
-/**
- * Code Implementation Module
- * Workflow for applying code changes and creating PRs
- */
-
-export { implementationStateSchema } from './state.js';
-export { buildImplementationGraph } from './graph.js';

package/dist/templates/code-implementation/state.js

@@ -1,14 +0,0 @@
-/**
- * Implementation Workflow State Schema
- */
-
-import { z } from 'zod';
-import { analysisStateSchema } from '../code-analysis/state.js';
-
-/**
- * Implementation workflow state schema (extends analysis)
- */
-export const implementationStateSchema = analysisStateSchema.extend({
-  branchName: z.string().optional().describe('Git branch name for changes'),
-  prTitle: z.string().optional().describe('Pull request title template'),
-});

package/templates/code-implementation/graph.js

@@ -1,35 +0,0 @@
-/**
- * Implementation Graph - Apply code changes AND create PR
- * 
- * Flow:
- * 1. setup - Clone repos and init git baseline
- * 2. implement_code - Apply changes, commit, push to branch
- * 3. generate_tests - Generate test cases
- * 4. create_pr - Create GitHub Pull Request
- * 5. finalize - Upload results to API
- */
-
-import { setupNode } from '../code-analysis/nodes/setup-node.js';
-import { implementCodeNode } from '../code-analysis/nodes/generate-code-node.js';
-import { generateTestCasesNode } from '../code-analysis/nodes/generate-test-cases-node.js';
-import { createPRNode } from '../code-analysis/nodes/create-pr-node.js';
-import { finalizeNode } from '../code-analysis/nodes/finalize-node.js';
-import { implementationStateSchema } from './state.js';
-
-export function buildImplementationGraph(graph) {
-  graph.setStateSchema(implementationStateSchema);
-  
-  graph
-    .addNode('setup', setupNode)                   // ← Reused from analysis!
-    .addNode('implement_code', implementCodeNode)  // ← Real changes + push
-    .addNode('generate_test_cases', generateTestCasesNode)  // ← Generate human-readable test specs
-    .addNode('create_pr', createPRNode)            // ← Create GitHub PR
-    .addNode('finalize', finalizeNode)             // ← Reused from analysis!
-    .setEntryPoint('setup')
-    .addEdge('setup', 'implement_code')
-    .addEdge('implement_code', 'generate_test_cases')
-    .addEdge('generate_test_cases', 'create_pr')
-    .addEdge('create_pr', 'finalize');
-  
-  return graph;
-}

package/templates/code-implementation/index.js

@@ -1,7 +0,0 @@
-/**
- * Code Implementation Module
- * Workflow for applying code changes and creating PRs
- */
-
-export { implementationStateSchema } from './state.js';
-export { buildImplementationGraph } from './graph.js';

package/templates/code-implementation/state.js

@@ -1,14 +0,0 @@
-/**
- * Implementation Workflow State Schema
- */
-
-import { z } from 'zod';
-import { analysisStateSchema } from '../code-analysis/state.js';
-
-/**
- * Implementation workflow state schema (extends analysis)
- */
-export const implementationStateSchema = analysisStateSchema.extend({
-  branchName: z.string().optional().describe('Git branch name for changes'),
-  prTitle: z.string().optional().describe('Pull request title template'),
-});