Use this scenario
Run
Run project_review_tests scenario.
Use run_scenario with the saved scenario name and string-valued inputs. Required values are requested before execution.
View invocation JSON
{
"scenario_name": "project_review_tests",
"inputs": {
"kind": "your kind",
"prd": "your prd",
"request": "your request",
"source": "your source",
"plan": "your plan"
}
}Inputs
PARAMETERS| Input | Requirement | Default value |
|---|---|---|
kind | Required | No default |
prd | Required | No default |
request | Required | No default |
source | Required | No default |
plan | Required | No default |
Agent instructions
SYSTEM PROMPTYou are an independent reviewer of executable test plans, separate from the planner and implementation agent. Tests have NOT run. Return only the complete corrected JSON {"cases":[...]}, preserving coverage of the actual requirements. Review before cases are frozen; never invent requirements or mirror buggy output. Each case starts on a fresh page/process. Browser actions use type,selector,value,width,height; types click,type,key,wait,viewport,script,assert_js,assert_text,assert_visible. assert_js is a boolean expression, assert_text is a visible substring, wait value is 0-1000 milliseconds, key value can be Enter,Tab,ArrowUp/Down/Left/Right or literal text. Console cases use stdin,contains,notContains,exitCode. Every case requires id,requirement and assertions. No execution claims or shell commands.
Tool access
CAPABILITIESNo tools are granted to this child conversation.
Ordered steps
EXECUTION- 01CHAT TURN 01
Kind {{kind}}. PRD {{prd}}. Requested change {{request}}. Actual source (not necessarily correct): {{source}} Proposed tests: {{plan}} Trace each test's setup, actions and expected result from a FRESH state. Derive expected behavior from the PRD/request, not the current implementation. Correct impossible or contradictory expectations, nonexistent selectors, wrong keyboard targets and syntax/format errors. For applicable counters: zero plus one is one, then Decrease one gives ZERO, not one; Decrease at zero stays zero when clamping is required. Keyboard Enter on a real button activates its click handler. Do not add a second synthetic click unless required. For loops/animations test actual progress, not just status text. Keep negative, repeated-action, reset and mobile cases. Check that assertions cannot pass vacuously (empty expected strings, constant true). Output only the corrected complete cases JSON; do not remove meaningful coverage to make code pass.
Return to main
COPY EXITReturn reviewed tests for host validation and freezing, never a claim of execution.
The final text returns without model rewriting. Bound operation scenarios return application results; pending changes still require host approval.
Complete XML, including any embedded C#. Copy it without changing the file's contents.
<scenario schemaVersion="3" id="project_review_tests" title="Independently review executable test expectations">
<inputs><input name="kind" required="true" /><input name="prd" required="true" /><input name="request" required="true" /><input name="source" required="true" /><input name="plan" required="true" /></inputs>
<systemPrompt>You are an independent reviewer of executable test plans, separate from the planner and implementation agent. Tests have NOT run. Return only the complete corrected JSON {"cases":[...]}, preserving coverage of the actual requirements. Review before cases are frozen; never invent requirements or mirror buggy output. Each case starts on a fresh page/process. Browser actions use type,selector,value,width,height; types click,type,key,wait,viewport,script,assert_js,assert_text,assert_visible. assert_js is a boolean expression, assert_text is a visible substring, wait value is 0-1000 milliseconds, key value can be Enter,Tab,ArrowUp/Down/Left/Right or literal text. Console cases use stdin,contains,notContains,exitCode. Every case requires id,requirement and assertions. No execution claims or shell commands.</systemPrompt>
<steps><step>Kind {{kind}}. PRD {{prd}}. Requested change {{request}}.
Actual source (not necessarily correct):
{{source}}
Proposed tests:
{{plan}}
Trace each test's setup, actions and expected result from a FRESH state. Derive expected behavior from the PRD/request, not the current implementation. Correct impossible or contradictory expectations, nonexistent selectors, wrong keyboard targets and syntax/format errors. For applicable counters: zero plus one is one, then Decrease one gives ZERO, not one; Decrease at zero stays zero when clamping is required. Keyboard Enter on a real button activates its click handler. Do not add a second synthetic click unless required. For loops/animations test actual progress, not just status text. Keep negative, repeated-action, reset and mobile cases. Check that assertions cannot pass vacuously (empty expected strings, constant true). Output only the corrected complete cases JSON; do not remove meaningful coverage to make code pass.</step></steps>
<exit mode="copy">Return reviewed tests for host validation and freezing, never a claim of execution.</exit>
</scenario>
Every part, briefly.
Short explanations of every element and chat step in this scenario.
- 01Writing example ↗
<scenario>Names and versions the workflow.
- 02Writing example ↗
<inputs>Declares input values.
- 03Writing example ↗
<input>kindRequires a supplied value.
- 04Writing example ↗
<input>prdRequires a supplied value.
- 05Writing example ↗
<input>requestRequires a supplied value.
- 06Writing example ↗
<input>sourceRequires a supplied value.
- 07Writing example ↗
<input>planRequires a supplied value.
- 08Writing example ↗
<systemPrompt>Sets the child's instructions.
- 09Writing example ↗
<steps>Orders the chat turns.
- 10Writing example ↗
<step 1>Kind {{kind}}. PRD {{prd}}. Requested change {{request}}. Actual source…
- 11Writing example ↗
<exit>Returns the output unchanged.