Skills google-agents-cli-eval
📦

google-agents-cli-eval

v0.5.1 Content revision r1 Medium Risk 🔑 Env variables⚙️ External commands🌐 Network access

Evaluate Google ADK Agents with agents-cli

Agent teams need repeatable evidence before changing prompts, tools, or deployment settings. This skill guides dataset design, metric selection, grading, comparison, and failure analysis for Google ADK agents.

Supports: Claude Codex Code(CC)
📊 71 Adequate

Install with my Agent

Copy this request to your Agent. It includes the canonical Skill page and manifest.

Agent request
Review the Skillstore skill "google-agents-cli-eval" from https://skillstore.io/skills/google-google-agents-cli-eval.md and its manifest at https://skillstore.io/api/skills/google-google-agents-cli-eval/manifest. Verify the artifact. You may proceed after verification, subject to the environment's own policy.

Your Agent should still show its plan and request any confirmation required by the security policy.

Agent-readable resources

Use these links when an AI agent, crawler, or script needs clean context instead of reading the full page.

Markdown detail GET /skills/google-google-agents-cli-eval.md Signed manifest GET /api/skills/google-google-agents-cli-eval/manifest Signed lockfile GET /api/skills/google-google-agents-cli-eval/lockfile

Test it

Using "google-agents-cli-eval". A user asks which metrics to use for a customer support agent.

Expected outcome:

  • Recommend task success for complete issue resolution.
  • Add tool use quality to verify case lookup and escalation calls.
  • Use hallucination and safety metrics for grounded, compliant responses.

Using "google-agents-cli-eval". A user reports that tool use quality dropped after a prompt change.

Expected outcome:

  • Inspect the latest grade report for failed rubric items.
  • Compare the failed trace with the previous passing trace.
  • Update tool descriptions or routing instructions, then rerun generate and grade.

Using "google-agents-cli-eval". A user wants to test an agent that handles image inputs.

Expected outcome:

  • Use multimodal dataset guidance from the reference file.
  • Select a vision-capable judge for image-aware scoring.
  • Avoid text-only metrics when the original media must be evaluated.

Security Audit

Medium Risk
v4 • 7/5/2026 Open versioned report

Most static findings are false positives caused by Markdown inline code, fenced examples, and documentation links rather than executable skill code. The remaining medium-risk issue is documented local custom metric execution with the CLI process privileges. No prompt injection or secret exfiltration intent was found.

6
Files scanned
1,248
Lines analyzed
0
Review items
0
False positives ignored

Confirmed security concerns (1)

Medium
SKILL.md:383-393
Local Custom Metric Code Execution
The guide states that code-based metrics run locally in process and that evaluate(instance) executes with the CLI process privileges. This is a legitimate feature, but untrusted custom metric code could read files, environment variables, or network resources available to the user.
The documented text explicitly says custom_function creates a CodeExecutionMetric and local execution uses the CLI privileges. The risk is user-initiated and disclosed, so severity is medium.

Risk Factors

🔑 Env variables (3)
references/builtin-tools-eval.md:141 references/builtin-tools-eval.md:141 references/builtin-tools-eval.md:142
⚙️ External commands (127)
SKILL.md:23 SKILL.md:25 SKILL.md:31 SKILL.md:32 SKILL.md:33 SKILL.md:34 SKILL.md:35 SKILL.md:45 SKILL.md:47 SKILL.md:51 SKILL.md:55 SKILL.md:57 SKILL.md:61 SKILL.md:63 SKILL.md:69 SKILL.md:71 SKILL.md:75 SKILL.md:86 SKILL.md:95 SKILL.md:96 SKILL.md:97 SKILL.md:98 SKILL.md:99 SKILL.md:100 SKILL.md:101 SKILL.md:103 SKILL.md:109 SKILL.md:113 SKILL.md:114 SKILL.md:115 SKILL.md:116 SKILL.md:117 SKILL.md:118 SKILL.md:119 SKILL.md:120 SKILL.md:122 SKILL.md:128 SKILL.md:130 SKILL.md:134-140 SKILL.md:140-142 SKILL.md:142-144 SKILL.md:144 SKILL.md:146-152 SKILL.md:152-160 SKILL.md:160-164 SKILL.md:164-166 SKILL.md:166 SKILL.md:168-170 SKILL.md:170-172 SKILL.md:172-174 SKILL.md:174-176 SKILL.md:176-178 SKILL.md:178-180 SKILL.md:180-182 SKILL.md:182 SKILL.md:184-194 SKILL.md:194-196 SKILL.md:196-198 SKILL.md:198 SKILL.md:200-212 SKILL.md:212-214 SKILL.md:214 SKILL.md:216-218 SKILL.md:218 SKILL.md:220-226 SKILL.md:226-228 SKILL.md:228 SKILL.md:230 SKILL.md:232-238 SKILL.md:238-244 SKILL.md:244 SKILL.md:246-247 SKILL.md:247 SKILL.md:249-255 SKILL.md:255-257 SKILL.md:257-276 SKILL.md:276-278 SKILL.md:278-280 SKILL.md:280-311 SKILL.md:311-315 SKILL.md:315 SKILL.md:317-344 SKILL.md:344-346 SKILL.md:346 SKILL.md:352 SKILL.md:354 SKILL.md:355 SKILL.md:359-379 SKILL.md:379-381 SKILL.md:381 SKILL.md:383 SKILL.md:385 SKILL.md:387-388 SKILL.md:388 SKILL.md:389 SKILL.md:391 SKILL.md:393 SKILL.md:404 SKILL.md:408 SKILL.md:410-416 SKILL.md:416-420 SKILL.md:420 SKILL.md:424 SKILL.md:426-428 SKILL.md:428-430 SKILL.md:430-432 SKILL.md:432 SKILL.md:434-436 SKILL.md:436-438 SKILL.md:438-449 SKILL.md:449-453 SKILL.md:453 SKILL.md:461-462 SKILL.md:462-463 SKILL.md:463-464 SKILL.md:464-465 SKILL.md:465 SKILL.md:473 SKILL.md:474-476 SKILL.md:476-477 SKILL.md:477-487 SKILL.md:487 SKILL.md:488-494 SKILL.md:494-495 SKILL.md:495-496 SKILL.md:496 SKILL.md:497-498
🌐 Network access (2)
SKILL.md:23 SKILL.md:103
Audited by: codex View Audit History →
Share & cite this report

Share the versioned assessment report, neutral badge, embed card, and citations. Skillstore reports evidence without deciding whether this Skill is safe.

Open versioned report
Security Assessment

Copy report link

https://skillstore.io/skills/google-google-agents-cli-eval/audits/4?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown badge

[![Skillstore security assessment](https://skillstore.io/badges/skills/google-google-agents-cli-eval/security.svg)](https://skillstore.io/skills/google-google-agents-cli-eval?utm_source=security_passport_badge)

HTML badge

<a href="https://skillstore.io/skills/google-google-agents-cli-eval?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/google-google-agents-cli-eval/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

Embed card

<iframe src="https://skillstore.io/embed/skills/google-google-agents-cli-eval.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>
Academic citations (APA · BibTeX · CFF)

APA citation

google. (2026). google-agents-cli-eval security audit report (audit version 4) [Author version 0.5.1]. Skillstore. https://skillstore.io/skills/google-google-agents-cli-eval/audits/4

BibTeX citation

@techreport{google-google-google-agents-cli-eval-2026, author = {google}, title = {google-agents-cli-eval security audit report (audit version 4)}, institution = {Skillstore}, year = {2026}, number = {4}, url = {https://skillstore.io/skills/google-google-agents-cli-eval/audits/4}, note = {Author version 0.5.1} }

CITATION.cff

cff-version: 1.2.0 message: "If you use this Skill, cite its author and this versioned security audit report." title: "google-agents-cli-eval security audit report (audit version 4)" version: "0.5.1" type: report authors: - name: "google" date-released: "2026-07-05" url: "https://skillstore.io/skills/google-google-agents-cli-eval/audits/4" identifiers: - type: other value: "skillstore:google-google-agents-cli-eval:audit:4" description: "Skillstore immutable audit report identifier"

Skillstore Score

Why this score Evidence Confidence: High
45
Architecture
100
Maintainability
87
Content
67
Community
100
Spec Compliance

What You Can Build

Validate prompt changes

Run generated traces and compare grade results before merging prompt or tool description updates.

Build evaluation coverage

Create datasets and select metrics that measure task success, tool behavior, safety, and hallucination risks.

Review cloud eval workflows

Use managed submit and results flows when local grading is too small for production-scale testing.

Try These Prompts

Choose metrics 
I am evaluating a [type] ADK agent. Help me choose built-in agents-cli eval metrics and explain what each metric will catch.
Draft a dataset plan 
Create an evaluation plan for these user goals: [goals]. Include case types, expected evidence, and common dataset mistakes to avoid.
Analyze failures 
Given these grade result symptoms: [symptoms]. Identify likely root causes and suggest agent or dataset changes before the next eval run.
Optimize the eval loop 
Design an advanced Quality Flywheel workflow for [agent]. Include dataset synthesis, metric selection, comparison gates, and when to use prompt optimization.

Best Practices

  • Start with a small dataset, fix failures, then expand coverage after passing cases are stable.
  • Use comparison runs to confirm a change improves the target metric without regressing other metrics.
  • Prefer remote sandbox execution for untrusted code-based custom metrics.

Avoid

  • Lowering thresholds to make an agent pass instead of fixing behavior.
  • Deleting flaky cases without diagnosing non-determinism or unclear rubrics.
  • Running expensive optimization loops before manual fixes and targeted eval evidence.

Frequently Asked Questions

Does this skill run evaluations by itself?
No. It guides the AI assistant and user through agents-cli eval workflows and commands.
Which tools does it support?
It is marked compatible with Claude, Codex, and Claude Code.
Does it replace official command help?
No. It tells users to check agents-cli help because flags can change.
Can it handle multimodal evaluations?
Yes. It includes guidance for image, audio, file, and video evaluation patterns.
Does it require cloud services?
Local generate and grade flows can run locally, but synthesize, submit, results, and some metrics use managed services.
What security care is needed?
Do not run untrusted local custom metric code. Use remote sandbox execution or review code before grading.

Developer Details

Author

google

License

MIT

Author version

v0.5.1

Skillstore revision

r1

Repository

https://github.com/google/agents-cli/tree/main/skills/google-agents-cli-eval/

Ref

30c73eac2afe762f6aa9c4553158769369d47351

Maintenance freshness

7/18/2026

Usage

5 downloads · 0 views

File structure

📁 references/

📄 builtin-tools-eval.md

📄 dataset_schema.md

📄 metrics-guide.md

📄 multimodal-eval.md

📄 user-simulation.md

📄 SKILL.md