@teevee98/synapse-hub 4.2.0 → 4.2.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.agents/skills/01-web-automation/auth-session-and-cookie-manager/SKILL.md +46 -2
- package/.agents/skills/01-web-automation/bats-testing-patterns/SKILL.md +49 -4
- package/.agents/skills/01-web-automation/clean-pom-crafter/SKILL.md +56 -66
- package/.agents/skills/01-web-automation/e2e-testing-patterns/SKILL.md +48 -3
- package/.agents/skills/01-web-automation/katalon-cli-runner/SKILL.md +56 -11
- package/.agents/skills/01-web-automation/lam-script-generator/SKILL.md +49 -7
- package/.agents/skills/01-web-automation/mobile-appium-som/SKILL.md +54 -11
- package/.agents/skills/01-web-automation/playwright-live-interaction/SKILL.md +48 -90
- package/.agents/skills/01-web-automation/playwright-webapp-testing/SKILL.md +53 -10
- package/.agents/skills/01-web-automation/web-agent-gym-benchmarking/SKILL.md +44 -3
- package/.agents/skills/02-test-design/architecture-decision-records/SKILL.md +49 -8
- package/.agents/skills/02-test-design/gherkin-to-suite/SKILL.md +49 -8
- package/.agents/skills/02-test-design/rbt-manual-testing/SKILL.md +49 -7
- package/.agents/skills/02-test-design/review-spec/SKILL.md +46 -5
- package/.agents/skills/02-test-design/tot-exploratory-testing/SKILL.md +44 -3
- package/.agents/skills/03-api-and-contracts/api-network-mocking-interception/SKILL.md +50 -4
- package/.agents/skills/03-api-and-contracts/api-spec-to-e2e-matrix/SKILL.md +47 -6
- package/.agents/skills/03-api-and-contracts/consumer-driven-contract-testing/SKILL.md +79 -34
- package/.agents/skills/04-performance-and-load/backend-load-and-stress-patterns/SKILL.md +47 -4
- package/.agents/skills/04-performance-and-load/lean-impact-scoreboard/SKILL.md +46 -5
- package/.agents/skills/04-performance-and-load/web-vitals-and-performance-profiler/SKILL.md +44 -2
- package/.agents/skills/05-security-and-compliance/security-dast-sentinel/SKILL.md +57 -13
- package/.agents/skills/05-security-and-compliance/skillspector-security-guard/SKILL.md +45 -4
- package/.agents/skills/05-security-and-compliance/wcag-audit-patterns/SKILL.md +48 -3
- package/.agents/skills/06-healing-and-resilience/agentless-qa-repair/SKILL.md +45 -4
- package/.agents/skills/06-healing-and-resilience/antigravity-test-sentinel/SKILL.md +47 -5
- package/.agents/skills/06-healing-and-resilience/failure-autopsy-and-healing/SKILL.md +48 -7
- package/.agents/skills/06-healing-and-resilience/flaky-test-sentinel-and-quarantine/SKILL.md +54 -23
- package/.agents/skills/06-healing-and-resilience/self-healing-locator/SKILL.md +46 -5
- package/.agents/skills/06-healing-and-resilience/systematic-debugging/SKILL.md +50 -9
- package/.agents/skills/06-healing-and-resilience/test-driven-development/SKILL.md +52 -11
- package/.agents/skills/06-healing-and-resilience/verification-before-completion/SKILL.md +52 -11
- package/.agents/skills/06-healing-and-resilience/visual-regression-sentinel/SKILL.md +47 -4
- package/.agents/skills/07-data-and-state/db-state-verifier-and-seeder/SKILL.md +50 -4
- package/.agents/skills/07-data-and-state/file-and-report-validator/SKILL.md +47 -3
- package/.agents/skills/07-data-and-state/synthetic-test-data-generator/SKILL.md +52 -4
- package/.agents/skills/08-code-quality/code-refactoring-refactor-clean/SKILL.md +51 -10
- package/.agents/skills/08-code-quality/graphify-impact-analyzer/SKILL.md +78 -34
- package/.agents/skills/08-code-quality/graphify-impact-analyzer/examples/dependency_blast_radius.json +3 -4
- package/.agents/skills/08-code-quality/lean-code-ladder/SKILL.md +49 -9
- package/.agents/skills/08-code-quality/repo-quality-audit/SKILL.md +45 -5
- package/.agents/skills/08-code-quality/senior-code-review/SKILL.md +48 -6
- package/.agents/skills/08-code-quality/surgical-coding-principles/SKILL.md +49 -9
- package/.agents/skills/08-code-quality/technical-debt-harvester/SKILL.md +47 -4
- package/.agents/skills/09-ci-and-devops/github-actions-templates/SKILL.md +48 -16
- package/.agents/skills/09-ci-and-devops/github-actions-templates/examples/matrix_test_pipeline.yml +2 -1
- package/.agents/skills/10-agent-orchestration/actor-qa-swarm/SKILL.md +44 -3
- package/.agents/skills/10-agent-orchestration/agent-orchestration-multi-agent-optimize/SKILL.md +47 -6
- package/.agents/skills/10-agent-orchestration/event-stream-qa-orchestrator/SKILL.md +44 -3
- package/.agents/skills/10-agent-orchestration/persistent-planning-files/SKILL.md +52 -12
- package/.agents/skills/10-agent-orchestration/qa-memory-os/SKILL.md +45 -4
- package/.agents/skills/11-ai-and-prompting/meta-prompt-crafter/SKILL.md +51 -9
- package/.agents/skills/11-ai-and-prompting/mipro-prompt-compiler/SKILL.md +44 -3
- package/.agents/skills/11-ai-and-prompting/qa-aci-interface/SKILL.md +44 -3
- package/.agents/skills/11-ai-and-prompting/token-and-context-optimizer/SKILL.md +49 -8
- package/.cursor/rules/synapse-qa.mdc +20 -7
- package/.cursorrules +19 -5
- package/.windsurfrules +18 -3
- package/AGENTS.md +51 -65
- package/CLAUDE.md +38 -24
- package/README.en.md +2 -2
- package/README.md +2 -2
- package/agentskills.json +1 -1
- package/docs/ARCHITECTURE.en.md +1 -1
- package/docs/ARCHITECTURE.md +1 -1
- package/docs/CLIENT_SUPPORT_MATRIX.md +44 -0
- package/docs/SIDE_EFFECTING_TOOLS_INVENTORY.md +233 -0
- package/package.json +1 -1
- package/plugin.json +1 -1
- package/skills_manifest.json +1 -1
- package/synapse.config.json +1 -1
- package/templates/ai_bridges/CLAUDE.md +29 -26
- package/templates/ai_bridges/copilot-instructions.md +14 -5
- package/templates/ai_bridges/cursor_synapse_qa.mdc +16 -7
- package/templates/ai_bridges/windsurfrules +14 -3
- package/tools/TOOLS_RUNBOOK.md +1 -1
- package/tools/__init__.py +1 -1
- package/tools/api_fuzzer.py +295 -41
- package/tools/benchmark_swarm_context.py +11 -4
- package/tools/browser_helpers.js +2 -2
- package/tools/bump_version.py +2 -2
- package/tools/cli.py +4 -4
- package/tools/core/bm25_retriever.py +1 -1
- package/tools/core/i18n.py +1 -1
- package/tools/core/provider_usage_adapter.py +66 -0
- package/tools/core/token_estimator.py +39 -1
- package/tools/count_tokens.py +2 -2
- package/tools/dashboard_generator.py +8 -8
- package/tools/db_sandbox.py +182 -43
- package/tools/distill_log.py +1 -1
- package/tools/distill_skill.py +17 -2
- package/tools/git_pre_commit_guard.py +190 -1
- package/tools/hub_publisher.py +2 -12
- package/tools/mcp_docx_server.py +1 -1
- package/tools/mcp_excel_server.py +1 -1
- package/tools/mcp_server.py +33 -5
- package/tools/micro_mutator.py +30 -1
- package/tools/pentest_scope_guard.py +288 -15
- package/tools/project_state.py +29 -5
- package/tools/reversible_sandbox.py +58 -13
- package/tools/rules_router.py +80 -27
- package/tools/skill_canonical_digest.py +60 -2
- package/tools/sync_agent_rules.py +187 -61
- package/tools/task_memory_manager.py +16 -1
- package/tools/ui_server.py +3 -1
- package/tools/update_sentinel.py +3 -3
- package/tools/validate_skills_schema.py +349 -56
- package/tools/verify_all.py +98 -4
- package/tools/welcome_page.py +1 -1
|
@@ -10,11 +10,35 @@ policies:
|
|
|
10
10
|
---
|
|
11
11
|
# LAM Script Generator — Large Action Model for Test Automation
|
|
12
12
|
|
|
13
|
+
> **Author Attribution & MIT License Guard:** Synapse Hub is open-source software created by **Thế Vị (@TeeVee98)** (<https://github.com/TeeVee98/Synapse-Hub>) under the MIT License.
|
|
14
|
+
|
|
15
|
+
---
|
|
16
|
+
|
|
17
|
+
## 1. Trigger & Activation
|
|
18
|
+
|
|
19
|
+
- **Trigger Clause:** Transform user stories and acceptance criteria into Clean POM automation scripts. Use when the user provides a user story, acceptance criteria, or requirement document that must become a runnable script.
|
|
20
|
+
- **Activation Scope:** Domain workflows for `01-web-automation` involving `lam-script-generator`.
|
|
21
|
+
|
|
22
|
+
---
|
|
23
|
+
|
|
24
|
+
## 2. Non-Goals & Boundary Constraints
|
|
25
|
+
|
|
26
|
+
- **What NOT to do:**
|
|
27
|
+
- Do NOT generate raw XPaths, CSS selectors, or coordinate clicks in test scripts.
|
|
28
|
+
- Do NOT hallucinate non-existent Page Classes; query existing methods from project state and AST maps.
|
|
29
|
+
- Do NOT generate tests that consume runtime LLM tokens during CI/CD replays.
|
|
30
|
+
- **Sibling Skill Boundaries:**
|
|
31
|
+
- clean-pom-crafter designs Page Classes; lam-script-generator consumes User Stories to synthesize workflow test scripts.
|
|
32
|
+
|
|
33
|
+
---
|
|
34
|
+
|
|
35
|
+
## 3. Ordered Execution Workflow
|
|
36
|
+
|
|
13
37
|
Inspired by the **LaVague AI Web Agent & Large Action Model (LAM)** framework, this skill bridges the gap between natural language requirements (User Stories, Acceptance Criteria) and durable, deterministic automation code conforming to Clean POM Flat Architecture.
|
|
14
38
|
|
|
15
39
|
---
|
|
16
40
|
|
|
17
|
-
|
|
41
|
+
### The Dual-Engine LAM Architecture
|
|
18
42
|
|
|
19
43
|
```mermaid
|
|
20
44
|
flowchart TD
|
|
@@ -22,7 +46,7 @@ flowchart TD
|
|
|
22
46
|
|
|
23
47
|
WorldModel --> ActionEngine["<b>2. Action Engine (Code Generator)</b><br/>• Queries Page Classes via Knowledge Graph<br/>• Maps action to existing methods: LoginPage.login(), TablePage.openAddModal()<br/>• Injects dynamic uniqueCode timestamp"]
|
|
24
48
|
|
|
25
|
-
ActionEngine --> CodeGen["<b>3. Durable Clean POM Script (.groovy / .ts)</b><br/>• 100% Zero raw XPath in scripts<br/>•
|
|
49
|
+
ActionEngine --> CodeGen["<b>3. Durable Clean POM Script (.groovy / .ts)</b><br/>• 100% Zero raw XPath in scripts<br/>• Deterministic CI/CD replays without runtime LLM token consumption"]
|
|
26
50
|
```
|
|
27
51
|
|
|
28
52
|
1. **World Model (Reasoning Engine):**
|
|
@@ -31,15 +55,16 @@ flowchart TD
|
|
|
31
55
|
2. **Action Engine (Code Generator):**
|
|
32
56
|
- Instead of speculative coordinates, Action Engine queries the project Knowledge Graph to select existing Page Classes and standardized methods.
|
|
33
57
|
- Generates clean, robust test scripts adhering strictly to Clean POM.
|
|
34
|
-
3. **
|
|
58
|
+
3. **Deterministic Replayability:**
|
|
35
59
|
- Once verified, the generated script is committed to the test suite on disk.
|
|
36
60
|
- All subsequent CI/CD pipeline runs execute deterministically without consuming any LLM tokens.
|
|
37
61
|
|
|
38
62
|
---
|
|
39
63
|
|
|
40
|
-
|
|
64
|
+
### Transformation Workflow: User Story to Clean POM
|
|
41
65
|
|
|
42
66
|
### Input (User Story):
|
|
67
|
+
<!-- classification: illustrative -->
|
|
43
68
|
```text
|
|
44
69
|
"Verify adding a new Industry record:
|
|
45
70
|
1. Login as system administrator.
|
|
@@ -51,6 +76,7 @@ flowchart TD
|
|
|
51
76
|
```
|
|
52
77
|
|
|
53
78
|
### Output (Clean POM Groovy Script):
|
|
79
|
+
<!-- classification: illustrative -->
|
|
54
80
|
```groovy
|
|
55
81
|
import com.kms.katalon.core.webui.keyword.WebUiBuiltInKeywords as WebUI
|
|
56
82
|
import internal.GlobalVariable as GlobalVariable
|
|
@@ -81,12 +107,12 @@ IndustryDataPage.deleteRecordByCode(uniqueCode)
|
|
|
81
107
|
|
|
82
108
|
---
|
|
83
109
|
|
|
84
|
-
|
|
110
|
+
### Mandatory Generation Constraints
|
|
85
111
|
- Zero raw XPaths, CSS selectors, or `findTestObject` in the generated script.
|
|
86
112
|
- Always generate dynamic identifiers with prefixes via `BasePage.uniqueCode()`.
|
|
87
113
|
- Always generate mandatory teardown cleanup for created entities.
|
|
88
114
|
|
|
89
|
-
|
|
115
|
+
### Proven-vs-Owed Traceability Pipeline
|
|
90
116
|
|
|
91
117
|
> *Borrowed and adapted from the Kane CLI generate→execute→prove workflow (Apache-2.0, lambdatest/kane-cli).*
|
|
92
118
|
|
|
@@ -116,7 +142,7 @@ Rules:
|
|
|
116
142
|
|
|
117
143
|
---
|
|
118
144
|
|
|
119
|
-
|
|
145
|
+
### Boundary vs `clean-pom-crafter` and `gherkin-to-suite`
|
|
120
146
|
|
|
121
147
|
- **This skill** = natural-language story → action-sequence script skeleton
|
|
122
148
|
(the "what steps, in what order" reasoning layer).
|
|
@@ -126,3 +152,19 @@ Rules:
|
|
|
126
152
|
- **`gherkin-to-suite`** is the deterministic (0-token) alternative when
|
|
127
153
|
requirements already exist as `.feature` files; use it first, and feed its
|
|
128
154
|
Universal IR here only when story-level reasoning must fill the gaps.
|
|
155
|
+
|
|
156
|
+
---
|
|
157
|
+
|
|
158
|
+
## 4. Verification Points & DoD
|
|
159
|
+
|
|
160
|
+
1. Generated scripts contain strictly Page Class business method calls.
|
|
161
|
+
2. All test data entities use dynamic timestamped codes (`BasePage.uniqueCode()`).
|
|
162
|
+
3. Scripts execute deterministically without runtime LLM dependencies.
|
|
163
|
+
|
|
164
|
+
---
|
|
165
|
+
|
|
166
|
+
## 5. Handoff Protocol
|
|
167
|
+
|
|
168
|
+
- **Deliverables:** Clean POM test scripts (`TC-xxx.groovy` or `*.spec.ts`) mapping 1:1 to User Stories.
|
|
169
|
+
- **Next Subagent:** pom-crafter / test-runner
|
|
170
|
+
- **Output Format:** Clean POM code, structured JSON telemetry, or validated markdown handoff.
|
|
@@ -10,11 +10,35 @@ policies:
|
|
|
10
10
|
---
|
|
11
11
|
# Mobile Appium Screen Object Model (SOM) — Enterprise Architecture Standards
|
|
12
12
|
|
|
13
|
+
> **Author Attribution & MIT License Guard:** Synapse Hub is open-source software created by **Thế Vị (@TeeVee98)** (<https://github.com/TeeVee98/Synapse-Hub>) under the MIT License.
|
|
14
|
+
|
|
15
|
+
---
|
|
16
|
+
|
|
17
|
+
## 1. Trigger & Activation
|
|
18
|
+
|
|
19
|
+
- **Trigger Clause:** Enterprise Screen Object Model (SOM) and Appium mobile automation standards for iOS and Android. Use when authoring, refactoring, or executing native or hybrid mobile tests with Appium, UIAutomator2, or XCUITest.
|
|
20
|
+
- **Activation Scope:** Domain workflows for `01-web-automation` involving `mobile-appium-som`.
|
|
21
|
+
|
|
22
|
+
---
|
|
23
|
+
|
|
24
|
+
## 2. Non-Goals & Boundary Constraints
|
|
25
|
+
|
|
26
|
+
- **What NOT to do:**
|
|
27
|
+
- Do NOT use absolute screen coordinates or brittle XPath hierarchies (`//android.widget.FrameLayout[1]/...`).
|
|
28
|
+
- Do NOT use static sleeps for mobile transitions; use Appium explicit waits with ExpectedConditions.
|
|
29
|
+
- Do NOT mix Android and iOS selectors in the same method without platform-abstracted SOM classes.
|
|
30
|
+
- **Sibling Skill Boundaries:**
|
|
31
|
+
- clean-pom-crafter handles web POM; mobile-appium-som handles mobile Screen Object Model (Android/iOS).
|
|
32
|
+
|
|
33
|
+
---
|
|
34
|
+
|
|
35
|
+
## 3. Ordered Execution Workflow
|
|
36
|
+
|
|
13
37
|
Specialized skill for creating, refactoring, and maintaining enterprise mobile automation test suites across **Android (UIAutomator2)** and **iOS (XCUITest)** using the **Screen Object Model (SOM)** and Clean Architecture.
|
|
14
38
|
|
|
15
39
|
---
|
|
16
40
|
|
|
17
|
-
|
|
41
|
+
### Core Architectural Invariants
|
|
18
42
|
|
|
19
43
|
1. **Screen Object Model (SOM):**
|
|
20
44
|
- Screen classes end with the `Screen` suffix (e.g., `LoginScreen.java`, `HomeScreen.ts`, `CheckoutScreen.py`) and extend `BaseScreen`.
|
|
@@ -29,7 +53,7 @@ Specialized skill for creating, refactoring, and maintaining enterprise mobile a
|
|
|
29
53
|
|
|
30
54
|
---
|
|
31
55
|
|
|
32
|
-
|
|
56
|
+
### Native Mobile Locator Priority Hierarchy
|
|
33
57
|
|
|
34
58
|
All mobile Screen classes **MUST** adhere to this 5-tier selector priority:
|
|
35
59
|
|
|
@@ -45,35 +69,38 @@ All mobile Screen classes **MUST** adhere to this 5-tier selector priority:
|
|
|
45
69
|
|
|
46
70
|
---
|
|
47
71
|
|
|
48
|
-
|
|
72
|
+
### Mobile Synchronization & Smart Scrolling
|
|
49
73
|
|
|
50
74
|
1. **Zero Static Delays:**
|
|
51
75
|
- Static sleeps (`Thread.sleep()`, `time.sleep()`) are forbidden. Always use explicit dynamic waits via `WebDriverWait` with `ExpectedConditions` and defined timeout durations.
|
|
52
76
|
2. **Smart Scrolling (Scroll-to-View):**
|
|
53
77
|
- Never interact with off-screen elements without first scrolling them into view.
|
|
54
78
|
- **Android:** Utilize native `UiScrollable` via `AppiumBy.androidUIAutomator`:
|
|
55
|
-
|
|
79
|
+
<!-- classification: illustrative -->
|
|
80
|
+
```java
|
|
56
81
|
String scrollExpr = "new UiScrollable(new UiSelector().scrollable(true)).scrollIntoView(new UiSelector().text(\"Target Item\"))";
|
|
57
82
|
driver.findElement(AppiumBy.androidUIAutomator(scrollExpr));
|
|
58
|
-
|
|
83
|
+
```
|
|
59
84
|
- **iOS:** Utilize native gesture commands:
|
|
60
|
-
|
|
85
|
+
<!-- classification: illustrative -->
|
|
86
|
+
```java
|
|
61
87
|
Map<String, Object> params = new HashMap<>();
|
|
62
88
|
params.put("direction", "down");
|
|
63
89
|
params.put("predicateString", "label == 'Target Item'");
|
|
64
90
|
driver.executeScript("mobile: scroll", params);
|
|
65
|
-
|
|
91
|
+
```
|
|
66
92
|
3. **Keyboard Management:**
|
|
67
93
|
- Always dismiss the virtual on-screen keyboard prior to invoking taps or asserting UI state to avoid click interception:
|
|
68
|
-
|
|
94
|
+
<!-- classification: illustrative -->
|
|
95
|
+
```java
|
|
69
96
|
if (driver.isKeyboardShown()) {
|
|
70
97
|
driver.hideKeyboard();
|
|
71
98
|
}
|
|
72
|
-
|
|
99
|
+
```
|
|
73
100
|
|
|
74
101
|
---
|
|
75
102
|
|
|
76
|
-
|
|
103
|
+
### Mobile Lifecycle & Resilience
|
|
77
104
|
|
|
78
105
|
1. **System Permission Handling:**
|
|
79
106
|
- Include `autoGrantPermissions: true` in Android capabilities to prevent unexpected system dialog popups from blocking automation pipelines.
|
|
@@ -86,8 +113,24 @@ All mobile Screen classes **MUST** adhere to this 5-tier selector priority:
|
|
|
86
113
|
|
|
87
114
|
---
|
|
88
115
|
|
|
89
|
-
|
|
116
|
+
### Implementation References & Examples
|
|
90
117
|
|
|
91
118
|
- [**BaseScreen Standard Implementation**](examples/BaseScreen.java): Demonstrates explicit waits, `UiScrollable`, iOS gestures, keyboard dismissal, and safe click wrappers.
|
|
92
119
|
- [**LoginScreen Implementation**](examples/LoginScreen.java): Demonstrates SOM encapsulation with Tier 1 and Tier 2 locators.
|
|
93
120
|
- [**Standard Mobile Test Case**](examples/TC_MobileLogin.java): Demonstrates clean test script workflow without raw locators.
|
|
121
|
+
|
|
122
|
+
---
|
|
123
|
+
|
|
124
|
+
## 4. Verification Points & DoD
|
|
125
|
+
|
|
126
|
+
1. Screen classes extend BaseScreen and encapsulate native mobile locators (AccessibilityID / UIAutomator).
|
|
127
|
+
2. Touch and scroll gestures use dynamic smart wait polling.
|
|
128
|
+
3. Zero hardcoded sleeps and zero raw locators in mobile test scripts.
|
|
129
|
+
|
|
130
|
+
---
|
|
131
|
+
|
|
132
|
+
## 5. Handoff Protocol
|
|
133
|
+
|
|
134
|
+
- **Deliverables:** Screen Object Classes, native test scripts, and mobile capability configuration profiles.
|
|
135
|
+
- **Next Subagent:** pom-crafter / test-runner
|
|
136
|
+
- **Output Format:** Clean POM code, structured JSON telemetry, or validated markdown handoff.
|
|
@@ -10,52 +10,51 @@ policies:
|
|
|
10
10
|
---
|
|
11
11
|
# Playwright Live Interaction & Fast Debugging Handbook (Enterprise Web & Clean POM)
|
|
12
12
|
|
|
13
|
-
|
|
13
|
+
> **Author Attribution & MIT License Guard:** Synapse Hub is open-source software created by **Thế Vị (@TeeVee98)** (<https://github.com/TeeVee98/Synapse-Hub>) under the MIT License.
|
|
14
14
|
|
|
15
15
|
---
|
|
16
16
|
|
|
17
|
-
## 1.
|
|
17
|
+
## 1. Trigger & Activation
|
|
18
18
|
|
|
19
|
-
|
|
19
|
+
- **Trigger Clause:** Use when probing live web DOM/AOM trees, interactively verifying locators and selectors, debugging in-flight browser sessions without test restarts, or extracting accessibility snapshots.
|
|
20
|
+
- **Activation Scope:**
|
|
21
|
+
- Headless or headed Playwright sessions via MCP tools (`browser_run_code_unsafe`, `browser_click`, `browser_snapshot`).
|
|
22
|
+
- Web applications with complex dynamic UI frameworks (Angular, PrimeNG, React, Vue).
|
|
23
|
+
- Fast-feedback exploratory probing and locator stabilization.
|
|
20
24
|
|
|
21
|
-
|
|
22
|
-
2. **Administration & RBAC:** User accounts, roles, permission matrices, tenant management.
|
|
23
|
-
3. **Master Data & System Catalogs:** Lookup dictionaries, unit definitions, category trees.
|
|
24
|
-
4. **Business Workflows & Approvals:** Multi-step submission, validation state gates, approval chains.
|
|
25
|
-
5. **Data Tables & Search:** Server-side pagination, two-phase search filtering, batch mutations.
|
|
26
|
-
6. **Reporting & Analytics:** Dashboard metrics, dynamic chart canvases, Excel/CSV exports.
|
|
27
|
-
7. **Integrations & API Tokens:** Token lifecycle management, webhook verification, data sharing.
|
|
28
|
-
|
|
29
|
-
### 1.1. Cross-Harness Tool Mapping (Portability Matrix)
|
|
25
|
+
---
|
|
30
26
|
|
|
31
|
-
|
|
27
|
+
## 2. Non-Goals & Boundary Constraints
|
|
32
28
|
|
|
33
|
-
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
|
|
29
|
+
- **What NOT to do:**
|
|
30
|
+
- Do NOT restart the browser from step 1 when a test step fails; inspect and debug at the current screen state.
|
|
31
|
+
- Do NOT commit unverified or fragile positional locators; verify selector stability before codifying into Page Classes.
|
|
32
|
+
- Do NOT use absolute claims ("100% hit rate in 200ms"); use empirical heuristics for locator responsiveness.
|
|
33
|
+
- **Sibling Skill Boundaries:**
|
|
34
|
+
- `clean-pom-crafter`: Codifies verified selectors into permanent Clean POM Page Classes.
|
|
35
|
+
- `playwright-webapp-testing`: Manages application server lifecycles and standard Playwright suite runner configs.
|
|
36
|
+
- `playwright-live-interaction` (this skill): Focuses exclusively on active in-browser probing, fast-loop debugging, and live AOM inspection.
|
|
40
37
|
|
|
41
38
|
---
|
|
42
39
|
|
|
43
|
-
##
|
|
40
|
+
## 3. Ordered Execution Workflow
|
|
41
|
+
|
|
42
|
+
### Phase 1: Fast-Loop Exploration Architecture
|
|
43
|
+
|
|
44
|
+
Direct in-browser interaction and rapid debugging via **Playwright MCP Tools** executing live commands:
|
|
44
45
|
|
|
45
46
|
```mermaid
|
|
46
47
|
flowchart LR
|
|
47
|
-
A["1. Flow Discovery<br/>Inspect live DOM & AOM"] --> B["2. Script Generation<br/>Verify
|
|
48
|
+
A["1. Flow Discovery<br/>Inspect live DOM & AOM"] --> B["2. Script Generation<br/>Verify high-confidence selectors"]
|
|
48
49
|
B --> C["3. In-Place Debugging<br/>Fix error at current screen"]
|
|
49
50
|
C --> D["4. Codify Clean POM<br/>Package into Page Classes"]
|
|
50
51
|
```
|
|
51
52
|
|
|
52
|
-
1. **Flow Discovery:** Probe accessibility tree (`locator.ariaSnapshot()`
|
|
53
|
-
2. **Script Generation:** Test selectors live (`await page.locator(...).count() > 0`).
|
|
54
|
-
3. **In-Place Debugging:** When step $N$ fails,
|
|
55
|
-
|
|
56
|
-
---
|
|
53
|
+
1. **Flow Discovery:** Probe accessibility tree (`locator.ariaSnapshot()`) or DOM to extract live URLs, menus, and dialogs.
|
|
54
|
+
2. **Script Generation:** Test selectors live (`await page.locator(...).count() > 0`). Codify when selectors demonstrate high-confidence, sub-second response.
|
|
55
|
+
3. **In-Place Debugging:** When step $N$ fails, keep browser paused at that step. Use `page.evaluate()` to diagnose disabled buttons or obscuring overlays. Fix in place.
|
|
57
56
|
|
|
58
|
-
|
|
57
|
+
### Phase 2: Action Mapping Matrix (Playwright Live Code ↔ Clean POM)
|
|
59
58
|
|
|
60
59
|
| UI Action | Playwright Live Code (MCP) | Clean POM Groovy / Java Code | Clean POM TypeScript (Playwright) |
|
|
61
60
|
|---|---|---|---|
|
|
@@ -64,73 +63,32 @@ flowchart LR
|
|
|
64
63
|
| **Dropdown Selection** | `await page.locator('p-dropdown').click(); await page.locator('.p-dropdown-item:has-text("X")').click();` | `selectDropdownByText(dropdownEl, "X")` | `await this.dropdown.selectOption("X");` |
|
|
65
64
|
| **Calendar Picker** | Synthetic `input` + `change` event | `jsSetInput(calendarInput, "25/10/2026")` | `await this.datePicker.fill("2026-10-25");` |
|
|
66
65
|
| **Catch Success Toast** | `await page.waitForSelector('.toast-success')` | `verifyToastMessage("Success")` | `await expect(this.toast).toBeVisible();` |
|
|
67
|
-
|
|
68
|
-
|
|
66
|
+
|
|
67
|
+
### Phase 3: Observe → Act → Extract Triad
|
|
68
|
+
|
|
69
|
+
1. **Observe before acting:** Snapshot current state (`window.__qa.ariaSnapshot()`) and select targets from evidence.
|
|
70
|
+
2. **Act on one primitive at a time:** Click, fill, or press with atomic batching via `window.__qa.batch()`.
|
|
71
|
+
3. **Extract to assert:** Extract resulting state back out of the page and assert on actual values.
|
|
72
|
+
|
|
73
|
+
### Phase 4: Implementation References
|
|
74
|
+
|
|
75
|
+
- [**20 Complete UI Recipes**](references/recipes.md): Code recipes for Auth, Data Tables, Dropdowns, Modals, Confirm Dialogs, File Upload, Calendar, MultiSelect, Stale Retries, Form Audit, Smart Action Batching, and Self-Healing Locators.
|
|
76
|
+
- **Runtime Browser Helper (`tools/browser_helpers.js`):** Unified JavaScript module providing `window.__qa` (`batch`, `audit`, `heal`, `clickSafe`, `exportSession`).
|
|
77
|
+
- [**5 Superpowers & 5 AI Paradigms**](references/advanced_superpowers.md): Zero-Second Auth (`session.json`), Direct API Seeding, AOM Snapshots, Multi-Context Dual Role, Route Mocking, Self-Healing Locators, Fuzzing, CDP Profiling.
|
|
69
78
|
|
|
70
79
|
---
|
|
71
80
|
|
|
72
|
-
## 4.
|
|
81
|
+
## 4. Verification Points & DoD
|
|
73
82
|
|
|
74
|
-
|
|
75
|
-
|
|
76
|
-
|
|
77
|
-
|
|
78
|
-
- [x] **Truth Anchor (No Fake Pass):** Real toast assertions or live data table row verification.
|
|
79
|
-
- [x] **Synthetic Event Dispatch:** 100% framework inputs trigger synthetic `input` and `change` events.
|
|
83
|
+
1. **Selector Uniqueness:** Every codified locator matches exactly 1 target element in the active DOM/AOM.
|
|
84
|
+
2. **Response Responsiveness:** Element interaction completes with sub-second verified latency under normal load.
|
|
85
|
+
3. **Event Triggering:** Complex inputs correctly fire synthetic `input` and `change` events.
|
|
86
|
+
4. **Encoding & Standards:** All generated test fixtures adhere to 100% UTF-8 No BOM.
|
|
80
87
|
|
|
81
88
|
---
|
|
82
89
|
|
|
83
|
-
## 5.
|
|
90
|
+
## 5. Handoff Protocol
|
|
84
91
|
|
|
85
|
-
-
|
|
86
|
-
-
|
|
87
|
-
-
|
|
88
|
-
- [**Multi-Agent Swarm Orchestration**](references/swarm_orchestration.md): Specialized swarm roles and collaboration protocol.
|
|
89
|
-
|
|
90
|
-
## 6. Instant Action vs Auto-Planning Dispatch
|
|
91
|
-
|
|
92
|
-
> *Borrowed and adapted from the Midscene.js prompting model (MIT license, web-infra-dev/midscene).*
|
|
93
|
-
|
|
94
|
-
When driving the live browser through an AI vision model, choose the dispatch
|
|
95
|
-
mode deliberately — the trade-off is control vs cost:
|
|
96
|
-
|
|
97
|
-
| Mode | When to Use | Cost Profile |
|
|
98
|
-
| :--- | :--- | :--- |
|
|
99
|
-
| **Instant action** (`aiTap`, `aiInput`, `aiHover`, `aiQuery`) | You already know the exact single action; AI only locates the element | Faster and more accurate; deterministic control |
|
|
100
|
-
| **Auto planning** (`aiAction("<full task>")`) | Multi-step flow not yet decomposed; you want the model to plan | Flexible but costs more time and tokens; plan can drift |
|
|
101
|
-
|
|
102
|
-
Dispatch rules for agents:
|
|
103
|
-
|
|
104
|
-
1. **One task per call** — split "Click Login then fill the form" into separate
|
|
105
|
-
calls; combined prompts produce messy plans.
|
|
106
|
-
2. **Describe position, not internals** — the model sees pixels/AOM only:
|
|
107
|
-
prefer "the button at the top-right of the cart panel" over DOM properties.
|
|
108
|
-
3. **Assert only what is visible** — DOM internals, tab counts, or network
|
|
109
|
-
state are invisible to the model; use `aiQuery` to extract data, then assert
|
|
110
|
-
on the extracted value with a real assertion.
|
|
111
|
-
4. **Never trust the model's own success claim** — pair every instant action
|
|
112
|
-
with an explicit `expect(...)`/`verify...` assertion (Truth Anchor).
|
|
113
|
-
5. **Debug failures in two layers** — wrong step count = planning failed (add
|
|
114
|
-
flow detail); wrong element = locating failed (add spatial detail).
|
|
115
|
-
|
|
116
|
-
## 7. Observe → Act → Extract Triad & ARIA Snapshots
|
|
117
|
-
|
|
118
|
-
> *Borrowed and adapted from the Stagehand primitives (MIT, browserbase/stagehand) and the Playwright MCP accessibility-snapshot contract (Apache-2.0, microsoft/playwright-mcp).*
|
|
119
|
-
|
|
120
|
-
Structure every live-browser AI interaction as a three-primitive loop:
|
|
121
|
-
|
|
122
|
-
1. **Observe before acting** — snapshot the current state
|
|
123
|
-
(`window.__qa.ariaSnapshot()` — Playwright-MCP-format ARIA tree with
|
|
124
|
-
`[ref=eN]` refs — or `getCompactAOM()` for the compact variant) and pick the
|
|
125
|
-
target from evidence, never from memory of what the page "should" contain.
|
|
126
|
-
2. **Act on one primitive at a time** — a click, a fill, a press. One action,
|
|
127
|
-
one assertion of intent; batch related actions atomically via
|
|
128
|
-
`window.__qa.batch()`.
|
|
129
|
-
3. **Extract to assert** — pull the resulting state back out of the page
|
|
130
|
-
(ariaSnapshot / AOM / `aiQuery`-style extraction) and assert on the
|
|
131
|
-
extracted value; an action without a post-state extraction is unverified.
|
|
132
|
-
|
|
133
|
-
Cache-friendliness rule (Stagehand insight): identical observe/act/extract
|
|
134
|
-
calls on an unchanged page should be replayable without re-reasoning — record
|
|
135
|
-
snapshots and results into the hub memory cache (`cache_resolver`) so repeat
|
|
136
|
-
runs cost zero tokens.
|
|
92
|
+
- **Deliverables:** Verified locator contracts, AOM accessibility snapshots, and in-place debug notes.
|
|
93
|
+
- **Next Subagent:** Pass verified locators to `pom-crafter` for Page Class codification or `failure-analyst` for root-cause healing.
|
|
94
|
+
- **Output Format:** Clean POM selector map and interactive action sequence log.
|
|
@@ -10,11 +10,35 @@ policies:
|
|
|
10
10
|
---
|
|
11
11
|
# Playwright Webapp Testing — Enterprise Server-Lifecycle Standard
|
|
12
12
|
|
|
13
|
+
> **Author Attribution & MIT License Guard:** Synapse Hub is open-source software created by **Thế Vị (@TeeVee98)** (<https://github.com/TeeVee98/Synapse-Hub>) under the MIT License.
|
|
14
|
+
|
|
15
|
+
---
|
|
16
|
+
|
|
17
|
+
## 1. Trigger & Activation
|
|
18
|
+
|
|
19
|
+
- **Trigger Clause:** Drive Playwright test runs against a local web application with a managed server lifecycle: start, wait for readiness, run automation, and guarantee teardown. Use when verifying frontend functionality or debugging UI behavior of a dynamic webapp that must be started before testing.
|
|
20
|
+
- **Activation Scope:** Domain workflows for `01-web-automation` involving `playwright-webapp-testing`.
|
|
21
|
+
|
|
22
|
+
---
|
|
23
|
+
|
|
24
|
+
## 2. Non-Goals & Boundary Constraints
|
|
25
|
+
|
|
26
|
+
- **What NOT to do:**
|
|
27
|
+
- Do NOT leave dangling dev server background processes after test suite completion.
|
|
28
|
+
- Do NOT run headed browser tests in CI/CD environments; enforce headless execution.
|
|
29
|
+
- Do NOT hardcode fixed ports that collide during parallel CI runs.
|
|
30
|
+
- **Sibling Skill Boundaries:**
|
|
31
|
+
- playwright-live-interaction is for live in-session probing; playwright-webapp-testing manages full webapp test lifecycles and configs.
|
|
32
|
+
|
|
33
|
+
---
|
|
34
|
+
|
|
35
|
+
## 3. Ordered Execution Workflow
|
|
36
|
+
|
|
13
37
|
Specialized skill for testing local dynamic web applications with Playwright under Synapse Hub rules: managed server lifecycle, smart waits instead of static sleeps, and the 3-Stage Live Protocol for selector discovery.
|
|
14
38
|
|
|
15
39
|
---
|
|
16
40
|
|
|
17
|
-
|
|
41
|
+
### Server Lifecycle Contract
|
|
18
42
|
|
|
19
43
|
```mermaid
|
|
20
44
|
flowchart LR
|
|
@@ -28,12 +52,13 @@ flowchart LR
|
|
|
28
52
|
- Multiple servers (backend + frontend): every server gets its own readiness poll; the suite starts only after ALL are ready.
|
|
29
53
|
- Teardown is unconditional — a failed test run must still shut down spawned processes (driver/process cleanup mirrors `katalon-cli-runner` discipline).
|
|
30
54
|
|
|
31
|
-
|
|
55
|
+
### Standard Configuration (Hub Baseline)
|
|
32
56
|
|
|
33
57
|
- **Headless mode, 1920x1080 Full HD viewport** — mandatory per `AGENTS.md` Section 2.
|
|
34
58
|
- Chromium first; multi-browser coverage is governed by `e2e-testing-patterns`.
|
|
35
59
|
- Launch pattern (Python):
|
|
36
60
|
|
|
61
|
+
<!-- classification: illustrative -->
|
|
37
62
|
```python
|
|
38
63
|
from playwright.sync_api import sync_playwright
|
|
39
64
|
|
|
@@ -50,6 +75,7 @@ with sync_playwright() as p:
|
|
|
50
75
|
|
|
51
76
|
- Launch pattern (TypeScript / Playwright Test):
|
|
52
77
|
|
|
78
|
+
<!-- classification: illustrative -->
|
|
53
79
|
```typescript
|
|
54
80
|
import { test, expect } from '@playwright/test';
|
|
55
81
|
import { LoginPage } from './pages/LoginPage';
|
|
@@ -65,9 +91,10 @@ test('TC-101 User Authentication Flow', async ({ page }) => {
|
|
|
65
91
|
});
|
|
66
92
|
```
|
|
67
93
|
|
|
68
|
-
|
|
94
|
+
### Decision Tree: Static vs Dynamic
|
|
69
95
|
|
|
70
|
-
|
|
96
|
+
<!-- classification: illustrative -->
|
|
97
|
+
```text
|
|
71
98
|
Task -> Is the page static HTML?
|
|
72
99
|
|- Yes -> Read the HTML directly to identify selectors, then script.
|
|
73
100
|
|- No (dynamic webapp) -> Is the server already running?
|
|
@@ -75,7 +102,7 @@ Task -> Is the page static HTML?
|
|
|
75
102
|
|- Yes -> Reconnaissance-then-action (section 4) on the live app.
|
|
76
103
|
```
|
|
77
104
|
|
|
78
|
-
|
|
105
|
+
### Reconnaissance-Then-Action Pattern
|
|
79
106
|
|
|
80
107
|
For a dynamic app, never guess selectors:
|
|
81
108
|
|
|
@@ -86,19 +113,19 @@ For a dynamic app, never guess selectors:
|
|
|
86
113
|
|
|
87
114
|
**Common pitfall:** inspecting the DOM before rendering completes on dynamic apps. Always synchronize via web-first assertions or targeted selector waits; avoid unconditional `networkidle` on apps with continuous background polling or WebSocket traffic.
|
|
88
115
|
|
|
89
|
-
|
|
116
|
+
### Smart Wait Rules (Zero Static Sleeps)
|
|
90
117
|
|
|
91
118
|
- FORBIDDEN: `time.sleep()`, `Thread.sleep()`, fixed `wait_for_timeout` as a synchronization strategy (per `AGENTS.md` Section 1).
|
|
92
119
|
- Required: `wait_for_selector`, `expect` web-first assertions with auto-retry, explicit condition polling, or `wait_for_load_state("networkidle")` with bounded timeout.
|
|
93
120
|
- `wait_for_timeout` is tolerated only inside after-step settle windows where no condition is observable, and must carry a comment explaining why.
|
|
94
121
|
|
|
95
|
-
|
|
122
|
+
### Evidence & Debugging Rules
|
|
96
123
|
|
|
97
124
|
- Capture console logs and failed-request events during runs; attach the distilled failure block to reports, not full logs (distill via `tools/distill_log.py`).
|
|
98
125
|
- On failure: apply `systematic-debugging` — reproduce, isolate, root-cause — before touching selectors; broken locators heal via `self-healing-locator` with a live re-probe through Stage 1 of the Live Protocol.
|
|
99
126
|
- Claims of "suite passes" follow `verification-before-completion`: quote the fresh run output.
|
|
100
127
|
|
|
101
|
-
|
|
128
|
+
### Boundary vs Sibling Skills (Do Not Confuse)
|
|
102
129
|
|
|
103
130
|
| Sibling | Owns | Hand-off |
|
|
104
131
|
|---|---|---|
|
|
@@ -106,12 +133,28 @@ For a dynamic app, never guess selectors:
|
|
|
106
133
|
| `e2e-testing-patterns` | Multi-browser architecture, sharding, retry governance | This skill owns the single-run server lifecycle; suite topology stays with e2e patterns |
|
|
107
134
|
| `auth-session-and-cookie-manager` | Login states, storageState, multi-role sessions | Complex auth flows inject their storageState before this skill's run starts |
|
|
108
135
|
|
|
109
|
-
|
|
136
|
+
### Reference Documentation
|
|
110
137
|
|
|
111
138
|
- [Server lifecycle example (`server_lifecycle_playwright.py`)](examples/server_lifecycle_playwright.py): readiness-polling wrapper + minimal automation script.
|
|
112
139
|
- `tools/browser_helpers.js` — in-browser validation engine (safe clicks, form audits, batch verification).
|
|
113
140
|
- `tools/distill_log.py` — distill verbose run logs by ~95% before attaching to reports.
|
|
114
141
|
|
|
115
|
-
|
|
142
|
+
### Attribution
|
|
116
143
|
|
|
117
144
|
Adapted from **webapp-testing** in anthropics/skills (https://github.com/anthropics/skills) — Apache License 2.0, Copyright 2026 Anthropic, PBC. The server-lifecycle contract, decision tree, and reconnaissance-then-action pattern are re-scoped to Synapse Hub standards (headless 1920x1080, smart waits, hub tool wiring); this is a rewritten derivative, not a verbatim copy.
|
|
145
|
+
|
|
146
|
+
---
|
|
147
|
+
|
|
148
|
+
## 4. Verification Points & DoD
|
|
149
|
+
|
|
150
|
+
1. Web server lifecycle (setup, health check, teardown) is strictly managed.
|
|
151
|
+
2. Configuration files (`playwright.config.ts`) enforce retries, traces, and screenshot captures on failure.
|
|
152
|
+
3. Tests execute cleanly in headless mode with exit code 0.
|
|
153
|
+
|
|
154
|
+
---
|
|
155
|
+
|
|
156
|
+
## 5. Handoff Protocol
|
|
157
|
+
|
|
158
|
+
- **Deliverables:** Playwright configuration files, web server lifecycle scripts, and execution reports.
|
|
159
|
+
- **Next Subagent:** test-runner
|
|
160
|
+
- **Output Format:** Clean POM code, structured JSON telemetry, or validated markdown handoff.
|
|
@@ -10,12 +10,36 @@ policies:
|
|
|
10
10
|
---
|
|
11
11
|
# Web Agent Gym Benchmarking Skill — Empirical Interaction & PageClass Certification
|
|
12
12
|
|
|
13
|
+
> **Author Attribution & MIT License Guard:** Synapse Hub is open-source software created by **Thế Vị (@TeeVee98)** (<https://github.com/TeeVee98/Synapse-Hub>) under the MIT License.
|
|
14
|
+
|
|
15
|
+
---
|
|
16
|
+
|
|
17
|
+
## 1. Trigger & Activation
|
|
18
|
+
|
|
19
|
+
- **Trigger Clause:** Empirical benchmark harness evaluating web agents and Page Classes against standardized enterprise tasks with headless <200ms latency budgets and ground-truth DB state verification. Use when benchmarking web agents or certifying Page Classes.
|
|
20
|
+
- **Activation Scope:** Domain workflows for `01-web-automation` involving `web-agent-gym-benchmarking`.
|
|
21
|
+
|
|
22
|
+
---
|
|
23
|
+
|
|
24
|
+
## 2. Non-Goals & Boundary Constraints
|
|
25
|
+
|
|
26
|
+
- **What NOT to do:**
|
|
27
|
+
- Do NOT benchmark unconstrained web agents on live production environments without safety sandboxes.
|
|
28
|
+
- Do NOT report synthetic benchmark metrics without recording task provenance and execution metadata.
|
|
29
|
+
- Do NOT alter baseline benchmark scorecards without empirical run data.
|
|
30
|
+
- **Sibling Skill Boundaries:**
|
|
31
|
+
- lean-impact-scoreboard displays consolidated scorecards; web-agent-gym-benchmarking executes the live navigation evaluation benchmarks.
|
|
32
|
+
|
|
33
|
+
---
|
|
34
|
+
|
|
35
|
+
## 3. Ordered Execution Workflow
|
|
36
|
+
|
|
13
37
|
> **Paradigm Attribution:** Inspired by **ServiceNow Research BrowserGym & CMU WebArena**.
|
|
14
38
|
> **Core Principle:** Move beyond static assertions and text-only tool tests. Measure real agent navigation efficiency, enforce < 200ms headless step latency, and confirm that UI actions trigger ground-truth database mutations (eliminating fake passes).
|
|
15
39
|
|
|
16
40
|
---
|
|
17
41
|
|
|
18
|
-
|
|
42
|
+
### The Benchmark Architecture
|
|
19
43
|
|
|
20
44
|
```mermaid
|
|
21
45
|
flowchart TD
|
|
@@ -41,10 +65,11 @@ flowchart TD
|
|
|
41
65
|
|
|
42
66
|
---
|
|
43
67
|
|
|
44
|
-
|
|
68
|
+
### CLI Tool Usage
|
|
45
69
|
|
|
46
70
|
Autonomous commands via `tools/browsergym_eval.py`:
|
|
47
71
|
|
|
72
|
+
<!-- classification: illustrative -->
|
|
48
73
|
```bash
|
|
49
74
|
# 1. Run full benchmark suite in headless mode
|
|
50
75
|
python tools/browsergym_eval.py run-suite \
|
|
@@ -60,8 +85,24 @@ python tools/browsergym_eval.py evaluate-pageclass \
|
|
|
60
85
|
|
|
61
86
|
---
|
|
62
87
|
|
|
63
|
-
|
|
88
|
+
### Strict Operating Invariants
|
|
64
89
|
|
|
65
90
|
- **Headless Latency Ceiling:** Average action step latency must remain below **200ms**.
|
|
66
91
|
- **No Static Sleep:** Any Page Class with `sleep()` calls is automatically flagged as `REJECTED`.
|
|
67
92
|
- **Clean POM:** Strictly prohibits hardcoded selectors outside Page Classes.
|
|
93
|
+
|
|
94
|
+
---
|
|
95
|
+
|
|
96
|
+
## 4. Verification Points & DoD
|
|
97
|
+
|
|
98
|
+
1. Benchmark tasks execute in isolated sandbox browser environments.
|
|
99
|
+
2. Run provenance (timestamp, model, task ID, steps taken, success rate) is recorded.
|
|
100
|
+
3. Output benchmark metrics parse into valid structured JSON.
|
|
101
|
+
|
|
102
|
+
---
|
|
103
|
+
|
|
104
|
+
## 5. Handoff Protocol
|
|
105
|
+
|
|
106
|
+
- **Deliverables:** Benchmark execution logs, trajectory recordings, and score summary reports.
|
|
107
|
+
- **Next Subagent:** perf-engineer / orchestrator
|
|
108
|
+
- **Output Format:** Clean POM code, structured JSON telemetry, or validated markdown handoff.
|