cxas-sim-eval

Name: cxas-sim-eval
Author: GoogleCloudPlatform

46installs64forks

How to install cxas-sim-eval?

googlecloudplatform/cxas-scrapi/cxas-sim-eval

$npx -y skills add GoogleCloudPlatform/cxas-scrapi --skill cxas-sim-eval

Installs into the current project.

Summary

Converts CXAS golden evaluations to SCRAPI SimulationEvals test cases. Use when generating high-level, goal-oriented test cases from turn-by-turn evaluation JSONs, and when enriching test expectations with inferred tool calls.

Files · 1

View on GitHub

SKILL.md

1	# CXAS Evaluation to Simulation Converter
2
3	This skill helps convert turn-by-turn CXAS golden evaluations into high-level, goal-oriented test cases for the SCRAPI `SimulationEvals` framework. It analyzes the agent's tools to enrich expectations with specific tool calls.
4
5	______________________________________________________________________
6
7	## Steps
8
9	### 1. Check Environment
10
11	Ensure `cxas_scrapi` is installed as a python package. You can check this by running:
12
13	```bash
14	python -c "import cxas_scrapi"
15	```
16
17	Ensure `gcloud` is authenticated properly:
18
19	```bash
20	gcloud auth list
21	```
22
23	If needed, login with:
24
25	```bash
26	gcloud auth login
27	```
28
29	### 2. Get App Name and Output Directory
30
31	> [!IMPORTANT]
32	> You MUST ask the user for the full resource name of the app/agent (e.g., `projects/.../locations/.../apps/...`) and the base output directory before proceeding with any execution steps.
33
34	Ask the user for these values.
35
36	### 3. Fetch Evaluations
37
38	Fetch the list of evaluations using the CES API. Save each evaluation as a JSON file named after its display name under `[output_dir]/golden_evals/`.
39
40	### 4. Fetch Tool Schemas
41
42	Fetch the full schemas for all tools available in the app and save them under `[output_dir]/tools/`.
43
44	### 5. Fetch Agent Tools Configuration
45
46	Fetch the list of tools and toolsets used by the agent and save the configuration (e.g., to `[output_dir]/agent_tools.json`).
47
48	### 6. Convert Evaluations
49
50	Run the conversion script (`convert_eval.py`) to process the fetched evaluations and save the converted test cases under `[output_dir]/sim_evals/`.
51
52	## Automation Scripts
53
54	Three scripts are available to automate the process:
55
56	### 1. Fetch Evaluations and Agent Config
57
58	`scripts/fetch_app_data.py`
59
60	Fetches evaluations and the list of tools used by the agent from the CES API.
61
62	Usage:
63
64	```bash
65	python .agents/skills/cxas-sim-eval/scripts/fetch_app_data.py \
66	--app-name "projects/.../locations/.../apps/..." \
67	--output-dir /path/to/output_directory
68	```
69
70	### 2. Fetch Tool Schemas
71
72	`scripts/fetch_tool_schemas.py`
73
74	Fetches the full schemas for all tools available in the app.
75
76	Usage:
77
78	```bash
79	python .agents/skills/cxas-sim-eval/scripts/fetch_tool_schemas.py \
80	--app-name "projects/.../locations/.../apps/..." \
81	--output-dir /path/to/output_directory
82	```
83
84	### 3. Convert Evaluations
85
86	`scripts/convert_eval.py`
87
88	Converts the fetched evaluations to simulation test cases, using the fetched tool schemas to infer expectations.
89
90	Usage:
91
92	```bash
93	python .agents/skills/cxas-sim-eval/scripts/convert_eval.py \
94	--output-dir /path/to/output_directory \
95	--parallelism 5
96	```
97
98	### 4. Run Evaluations
99
100	`scripts/run_evals.py`
101
102	Runs the simulation evaluations, logs raw results, and generates a combined HTML report.
103
104	Cognitive Diagnostics Analysis:
105	If the agent has the `intercept_and_score_reasoning` tool enabled, this script will automatically extract and analyze the agent's internal monologue for failed evaluations. It detects issues like overthinking, hesitation, and backtracking. Furthermore, it correlates these diagnostics with the agent's instructions to generate actionable suggestions for improvement directly in the HTML report.
106
107	Usage:
108
109	```bash
110	python .agents/skills/cxas-sim-eval/scripts/run_evals.py \
111	--app-name "projects/.../locations/.../apps/..." \
112	--output-dir /path/to/output_directory \
113	--parallelism 5 \
114	--start-index 0 \
115	--end-index 10
116	```
117
118	## Interpreting Cognitive Diagnostics
119
120	When running evaluations with the `intercept_and_score_reasoning` tool enabled, the system extracts diagnostics to help you identify issues in agent reasoning.
121
122	### Key Signals
123
124	1. Overthinking (Verbosity)
125
126	- Symptom: Internal monologue exceeds 350 or 600 characters.
127	- Meaning: The agent is struggling to process complex or circular instructions.
128	- Fix: Simplify instructions. Break down complex tasks into smaller, linear steps.
129
130	1. Hedging
131
132	- Symptom: Use of words like "might be", "guess", "unsure", "assume".
133	- Meaning: The agent is uncertain about its next action, often due to missing edge case handling.
134	- Fix: Add explicit instructions for the scenario the agent is unsure about.
135
136	1. Backtracking
137
138	- Symptom: Use of words like "wait", "actually", "on second thought".
139	- Meaning: The agent is abandoning a plan mid-turn or correcting itself, indicating unclear triggers.
140	- Fix: Clarify triggers and state transitions in instructions.

People who liked this also viewed

6 picks

Type

cxas-sim-eval

46installs64forks

View on GitHub

How to install cxas-sim-eval?

googlecloudplatform/cxas-scrapi/cxas-sim-eval

$npx -y skills add GoogleCloudPlatform/cxas-scrapi --skill cxas-sim-eval

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# CXAS Evaluation to Simulation Converter
2
3	This skill helps convert turn-by-turn CXAS golden evaluations into high-level, goal-oriented test cases for the SCRAPI `SimulationEvals` framework. It analyzes the agent's tools to enrich expectations with specific tool calls.
4
5	______________________________________________________________________
6
7	## Steps
8
9	### 1. Check Environment
10
11	Ensure `cxas_scrapi` is installed as a python package. You can check this by running:
12
13	```bash
14	python -c "import cxas_scrapi"
15	```
16
17	Ensure `gcloud` is authenticated properly:
18
19	```bash
20	gcloud auth list
21	```
22
23	If needed, login with:
24
25	```bash
26	gcloud auth login
27	```
28
29	### 2. Get App Name and Output Directory
30
31	> [!IMPORTANT]
32	> You MUST ask the user for the full resource name of the app/agent (e.g., `projects/.../locations/.../apps/...`) and the base output directory before proceeding with any execution steps.
33
34	Ask the user for these values.
35
36	### 3. Fetch Evaluations
37
38	Fetch the list of evaluations using the CES API. Save each evaluation as a JSON file named after its display name under `[output_dir]/golden_evals/`.
39
40	### 4. Fetch Tool Schemas
41
42	Fetch the full schemas for all tools available in the app and save them under `[output_dir]/tools/`.
43
44	### 5. Fetch Agent Tools Configuration
45
46	Fetch the list of tools and toolsets used by the agent and save the configuration (e.g., to `[output_dir]/agent_tools.json`).
47
48	### 6. Convert Evaluations
49
50	Run the conversion script (`convert_eval.py`) to process the fetched evaluations and save the converted test cases under `[output_dir]/sim_evals/`.
51
52	## Automation Scripts
53
54	Three scripts are available to automate the process:
55
56	### 1. Fetch Evaluations and Agent Config
57
58	`scripts/fetch_app_data.py`
59
60	Fetches evaluations and the list of tools used by the agent from the CES API.
61
62	Usage:
63
64	```bash
65	python .agents/skills/cxas-sim-eval/scripts/fetch_app_data.py \
66	--app-name "projects/.../locations/.../apps/..." \
67	--output-dir /path/to/output_directory
68	```
69
70	### 2. Fetch Tool Schemas
71
72	`scripts/fetch_tool_schemas.py`
73
74	Fetches the full schemas for all tools available in the app.
75
76	Usage:
77
78	```bash
79	python .agents/skills/cxas-sim-eval/scripts/fetch_tool_schemas.py \
80	--app-name "projects/.../locations/.../apps/..." \
81	--output-dir /path/to/output_directory
82	```
83
84	### 3. Convert Evaluations
85
86	`scripts/convert_eval.py`
87
88	Converts the fetched evaluations to simulation test cases, using the fetched tool schemas to infer expectations.
89
90	Usage:
91
92	```bash
93	python .agents/skills/cxas-sim-eval/scripts/convert_eval.py \
94	--output-dir /path/to/output_directory \
95	--parallelism 5
96	```
97
98	### 4. Run Evaluations
99
100	`scripts/run_evals.py`
101
102	Runs the simulation evaluations, logs raw results, and generates a combined HTML report.
103
104	Cognitive Diagnostics Analysis:
105	If the agent has the `intercept_and_score_reasoning` tool enabled, this script will automatically extract and analyze the agent's internal monologue for failed evaluations. It detects issues like overthinking, hesitation, and backtracking. Furthermore, it correlates these diagnostics with the agent's instructions to generate actionable suggestions for improvement directly in the HTML report.
106
107	Usage:
108
109	```bash
110	python .agents/skills/cxas-sim-eval/scripts/run_evals.py \
111	--app-name "projects/.../locations/.../apps/..." \
112	--output-dir /path/to/output_directory \
113	--parallelism 5 \
114	--start-index 0 \
115	--end-index 10
116	```
117
118	## Interpreting Cognitive Diagnostics
119
120	When running evaluations with the `intercept_and_score_reasoning` tool enabled, the system extracts diagnostics to help you identify issues in agent reasoning.
121
122	### Key Signals
123
124	1. Overthinking (Verbosity)
125
126	- Symptom: Internal monologue exceeds 350 or 600 characters.
127	- Meaning: The agent is struggling to process complex or circular instructions.
128	- Fix: Simplify instructions. Break down complex tasks into smaller, linear steps.
129
130	1. Hedging
131
132	- Symptom: Use of words like "might be", "guess", "unsure", "assume".
133	- Meaning: The agent is uncertain about its next action, often due to missing edge case handling.
134	- Fix: Add explicit instructions for the scenario the agent is unsure about.
135
136	1. Backtracking
137
138	- Symptom: Use of words like "wait", "actually", "on second thought".
139	- Meaning: The agent is abandoning a plan mid-turn or correcting itself, indicating unclear triggers.
140	- Fix: Clarify triggers and state transitions in instructions.

People who liked this also viewed

6 picks

Type