Alicloud Ai Entry Modelstudio Test

Run a minimal test matrix for the Model Studio skills that exist in this repo, including image/video/audio, realtime speech, omni, visual reasoning, embeddin...

Security Scan

The skill's instructions are generally consistent with a repo-level test runner, but the metadata omits required credentials/config paths and the runtime instructions ask the operator to run networked SDK calls that could be mutating or leak identifiers—review before running.

Purpose & Capability

The skill's stated purpose—run a minimal test matrix across many Model Studio sub-skills—matches the SKILL.md steps (open sub-skill SKILL.md, run one request per sub-skill). However the manifest metadata does not declare the credential/config needs that the instructions require (see environment_proportionality).

Instruction Scope

Runtime instructions tell the operator to install the 'dashscope' SDK, set DASHSCOPE_API_KEY or use ~/.alibabacloud/credentials, read many sub-skill SKILL.md files, run SDK/API calls, and save evidence including region/resource IDs/time ranges. That is expected for testing, but it also means tests may access real cloud resources, be mutating (some operations may create or modify content), and will record potentially sensitive identifiers into repository files—there is no built-in safeguard to enforce read-only testing.

Install Mechanism

This is instruction-only (no install spec). SKILL.md requires creating a venv and pip-installing 'dashscope'. That is a common pattern, but the package name is not validated in the metadata (no lockfile or vetted source). Installing arbitrary PyPI packages carries moderate risk—verify the package provenance before pip install.

Credentials

The SKILL.md explicitly requires DASHSCOPE_API_KEY (or ~/.alibabacloud/credentials) but the registry metadata lists no required env vars or config paths. This mismatch is material: the skill needs cloud credentials to function but does not declare them, which can mislead reviewers and automated gates.

Persistence & Privilege

always is false and the skill is instruction-only with no install spec that persists code or modifies other skills/config. It does instruct saving test outputs under an output/ path in the repo, which is normal for test artifacts.

What to consider before installing

Before running: (1) Expect to provide an Alibaba Cloud API key (DASHSCOPE_API_KEY) or credentials file—the registry metadata does not declare this, so add or verify it. (2) Run tests in an isolated environment (fresh venv, dedicated or limited-scope API key) so any accidental mutating calls or credentials are confined. (3) Inspect each referenced sub-skill's SKILL.md to confirm the exact API calls and whether they are read-only; some capabilities (video/editing/voice cloning) may create resources or be billable. (4) Verify the 'dashscope' PyPI package/project identity before pip install (ensure it's from the official source). (5) Avoid committing output files that include region/resource IDs, API keys, or other sensitive identifiers; sanitize or store evidence securely. (6) Consider updating the skill metadata to declare DASHSCOPE_API_KEY and the ~/.alibabacloud/credentials config path so automated reviewers and users are aware of the credential requirement.

Category: task

Model Studio Skills Minimal Test

Run minimal validation for currently available Model Studio skills in this repo and record results.

Prerequisites

• Install SDK (virtual environment recommended to avoid PEP 668 restrictions):

python3 -m venv .venv
. .venv/bin/activate
python -m pip install dashscope

• Configure DASHSCOPE_API_KEY (environment variable preferred; or dashscope_api_key in ~/.alibabacloud/credentials).

Test Matrix (currently supported)

1. Text-to-image → skills/ai/image/alicloud-ai-image-qwen-image/
2. Image editing → skills/ai/image/alicloud-ai-image-qwen-image-edit/
3. Text-to-video / Image-to-video (i2v) → skills/ai/video/alicloud-ai-video-wan-video/
4. Reference-to-video (r2v) → skills/ai/video/alicloud-ai-video-wan-r2v/
5. TTS → skills/ai/audio/alicloud-ai-audio-tts/
6. ASR transcription (non-realtime) → skills/ai/audio/alicloud-ai-audio-asr/
7. Realtime ASR → skills/ai/audio/alicloud-ai-audio-asr-realtime/
8. Realtime TTS → skills/ai/audio/alicloud-ai-audio-tts-realtime/
9. Live speech translation → skills/ai/audio/alicloud-ai-audio-livetranslate/
10. CosyVoice voice clone → skills/ai/audio/alicloud-ai-audio-cosyvoice-voice-clone/
11. CosyVoice voice design → skills/ai/audio/alicloud-ai-audio-cosyvoice-voice-design/
12. Voice clone → skills/ai/audio/alicloud-ai-audio-tts-voice-clone/
13. Voice design → skills/ai/audio/alicloud-ai-audio-tts-voice-design/
14. Omni multimodal → skills/ai/multimodal/alicloud-ai-multimodal-qwen-omni/
15. Visual reasoning → skills/ai/multimodal/alicloud-ai-multimodal-qvq/
16. Text embedding → skills/ai/search/alicloud-ai-search-text-embedding/
17. Rerank → skills/ai/search/alicloud-ai-search-rerank/
18. Video editing → skills/ai/video/alicloud-ai-video-wan-edit/

If new capability tests are needed, create corresponding skill first (use skills/ai/misc/alicloud-ai-misc-crawl-and-skill/ to refresh model list).

Minimal Flow Per Capability

1. Open target sub-skill directory and read SKILL.md.
2. Choose one minimal input example and recommended model.
3. Run SDK call or script.
4. Record model, request summary, response summary, duration, and status.

Result Template

Save as output/alicloud-ai-entry-modelstudio-test-results.md:

# Model Studio Skill Test Results

- Date: YYYY-MM-DD
- Environment: region / API_BASE / auth method

| Capability | Sub-skill | Model | Request summary | Result summary | Status | Notes |
| --- | --- | --- | --- | --- | --- | --- |
| Text-to-image | alicloud-ai-image-qwen-image | <model-id> | ... | ... | pass/fail | ... |
| Image editing | alicloud-ai-image-qwen-image-edit | <model-id> | ... | ... | pass/fail | ... |
| Image-to-video (i2v) | alicloud-ai-video-wan-video | <model-id> | ... | ... | pass/fail | ... |
| Reference-to-video (r2v) | alicloud-ai-video-wan-r2v | <model-id> | ... | ... | pass/fail | ... |
| TTS | alicloud-ai-audio-tts | <model-id> | ... | ... | pass/fail | ... |
| ASR (non-realtime) | alicloud-ai-audio-asr | <model-id> | ... | ... | pass/fail | ... |
| Realtime ASR | alicloud-ai-audio-asr-realtime | <model-id> | ... | ... | pass/fail | ... |
| Realtime TTS | alicloud-ai-audio-tts-realtime | <model-id> | ... | ... | pass/fail | ... |
| Live speech translation | alicloud-ai-audio-livetranslate | <model-id> | ... | ... | pass/fail | ... |
| CosyVoice voice clone | alicloud-ai-audio-cosyvoice-voice-clone | <model-id> | ... | ... | pass/fail | ... |
| CosyVoice voice design | alicloud-ai-audio-cosyvoice-voice-design | <model-id> | ... | ... | pass/fail | ... |
| Voice clone | alicloud-ai-audio-tts-voice-clone | <model-id> | ... | ... | pass/fail | ... |
| Voice design | alicloud-ai-audio-tts-voice-design | <model-id> | ... | ... | pass/fail | ... |
| Omni multimodal | alicloud-ai-multimodal-qwen-omni | <model-id> | ... | ... | pass/fail | ... |
| Visual reasoning | alicloud-ai-multimodal-qvq | <model-id> | ... | ... | pass/fail | ... |
| Text embedding | alicloud-ai-search-text-embedding | <model-id> | ... | ... | pass/fail | ... |
| Rerank | alicloud-ai-search-rerank | <model-id> | ... | ... | pass/fail | ... |
| Video editing | alicloud-ai-video-wan-edit | <model-id> | ... | ... | pass/fail | ... |

Failure Handling

• If parameters are unclear, check target sub-skill SKILL.md or references/*.md.
• If model is unavailable, refresh model list and retry.
• For auth issues, verify DASHSCOPE_API_KEY (env var or ~/.alibabacloud/credentials).

Validation

mkdir -p output/alicloud-ai-entry-modelstudio-test
echo "validation_placeholder" > output/alicloud-ai-entry-modelstudio-test/validate.txt

Pass criteria: command exits 0 and output/alicloud-ai-entry-modelstudio-test/validate.txt is generated.

Output And Evidence

• Save artifacts, command outputs, and API response summaries under output/alicloud-ai-entry-modelstudio-test/.
• Include key parameters (region/resource id/time range) in evidence files for reproducibility.

Workflow

1. Confirm user intent, region, identifiers, and whether the operation is read-only or mutating.
2. Run one minimal read-only query first to verify connectivity and permissions.
3. Execute the target operation with explicit parameters and bounded scope.
4. Verify results and save output/evidence files.

References

• Source list: references/sources.md