You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: agent/resources/skills/testar-cli/SKILL.md
+18-4Lines changed: 18 additions & 4 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -38,7 +38,8 @@ Treat the CLI distribution as an operational environment, not as a source-code w
38
38
-`executeAction click <semanticText>`
39
39
-`executeAction type <semanticText> <inputText>`
40
40
-`executeAction select <semanticText> <value>`
41
-
-`stopSession`
41
+
-`stopSession LLM_COMPLETE <reason>`
42
+
-`stopSession LLM_INVALID <reason>`
42
43
-`shutdownDaemon`
43
44
44
45
## Semantic action matching
@@ -68,6 +69,19 @@ Use short meaningful text that is likely to appear in the action description, fo
68
69
6. Stop the session when the goal is finished.
69
70
7. Shutdown the daemon after stopping the session.
70
71
72
+
## Test goal verdicts
73
+
74
+
Agent CLI executions must finish each test goal with exactly one explicit verdict command:
75
+
76
+
-`stopSession LLM_COMPLETE <reason>`
77
+
-`stopSession LLM_INVALID <reason>`
78
+
79
+
Use `LLM_COMPLETE` only when the requested test goal was completed and the expected result was verified with CLI evidence.
80
+
81
+
Use `LLM_INVALID` when the expected result is not verified, a bug or invalid state is observed, the goal cannot be completed, or the available evidence is not reliable enough to declare completion.
82
+
83
+
The `<reason>` must briefly explain the final decision using observed CLI evidence.
84
+
71
85
## Resilience and retry behavior rules
72
86
73
87
Some SUTs may need time to start, load screens, populate state information, or expose derived actions after an action is executed.
@@ -100,7 +114,7 @@ Workspace-driven Windows testing:
100
114
-`testar-cli.bat getDerivedActions`
101
115
-`testar-cli.bat executeAction click Open`
102
116
-`testar-cli.bat executeAction type Editor Writing_text_in_Notepad`
103
-
-`testar-cli.bat stopSession`
117
+
-`testar-cli.bat stopSession LLM_COMPLETE "Notepad opened and text was entered successfully."`
Copy file name to clipboardExpand all lines: docs/architecture_distribution_workspace.md
+30-1Lines changed: 30 additions & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -142,7 +142,36 @@ This prevents CLI from failing only because unrelated scriptless-only Java files
142
142
143
143
Oracle composer entries are scriptless-only.
144
144
145
-
CLI agent runs should use the agent as the oracle for LLM-complete verdicts, such as `LLM_COMPLETE_VALID` and `LLM_COMPLETE_INVALID`.
145
+
CLI agent runs should use the agent as the oracle for LLM verdicts, such as `LLM_COMPLETE` and `LLM_INVALID`.
146
+
147
+
## CLI Session Verdict Contract
148
+
149
+
`stopSession` is the CLI session finalization command.
150
+
151
+
Manual CLI execution may use:
152
+
153
+
```text
154
+
stopSession
155
+
```
156
+
157
+
Plain `stopSession` defaults the final session verdict to `OK`.
158
+
159
+
Agent CLI execution must finish each test goal with an explicit LLM verdict:
160
+
161
+
```text
162
+
stopSession LLM_COMPLETE <reason>
163
+
stopSession LLM_INVALID <reason>
164
+
```
165
+
166
+
`LLM_COMPLETE` means the agent completed the requested test goal and verified the expected result with CLI evidence.
167
+
168
+
`LLM_INVALID` means the agent reached an invalid result, found a bug, could not verify the expected result, or could not complete the goal with reliable CLI evidence.
169
+
170
+
The reason is free text and should explain the final decision.
171
+
172
+
If reporting is enabled, the final CLI verdict must be written to the generated report artifacts.
173
+
174
+
The `testar-cli` skill documentation is the operational source that tells agents how to use this contract.
Copy file name to clipboardExpand all lines: docs/webstudio/WEBSTUDIO_FUNCTIONAL_SPEC.md
+21Lines changed: 21 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -360,6 +360,27 @@ Dialog actions:
360
360
-`Discard`: restore the last persisted Agent CLI settings in the CLI panel, then continue the pending action
361
361
-`Cancel`: keep the user in CLI mode and abort the pending action
362
362
363
+
### CLI Verdict Finalization
364
+
365
+
Manual CLI execution may stop a session with plain `stopSession`.
366
+
367
+
Plain `stopSession` must finalize the session with an `OK` verdict.
368
+
369
+
Agent CLI execution must stop a session with an explicit LLM verdict:
370
+
371
+
-`stopSession LLM_COMPLETE <reason>`
372
+
-`stopSession LLM_INVALID <reason>`
373
+
374
+
`LLM_COMPLETE` means the agent completed the test goal and verified the expected result.
375
+
376
+
`LLM_INVALID` means the agent found an invalid result, could not verify the expected result, or could not complete the goal with reliable evidence.
377
+
378
+
WebStudio must include this finalization rule in the Agent CLI prompt contract.
379
+
380
+
If an Agent CLI execution ends without an explicit LLM verdict, WebStudio should treat the execution as incomplete or invalid and surface that status to the user.
381
+
382
+
When CLI reporting is enabled, generated report artifacts must include the final CLI verdict and reason.
383
+
363
384
## Test Results
364
385
365
386
The Test Results page supports inspection of output folders from the shared TESTAR distribution.
0 commit comments