Back to skill

Security audit

huawei-cloud-cloudrobo-asset

Security checks for vulnerabilities and agentic risk

Overview

Tests described as safe previews can delete real cloud assets, and the setup instructions document insecure connection settings.

Review before installing. Avoid running the bundled SDK tests with production credentials or existing asset IDs until their preview behavior is fixed. Use disposable resources and narrowly scoped credentials, enable CLOUDROBO_VERIFY_SSL=true, and confirm deletion targets and upload folders. The lifecycle diagram's cleanup step remains subject to SKILL.md's confirmation requirement, and folder uploads are explicitly disclosed. Scanner coverage gaps were manually investigated; the external SDK's runtime behavior was not tested.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/test-cli-commands.sh
Finding

SDK dry-run tests perform unconfirmed mutations and deletions

Content
View full analysis

Impact

Existing assets and versions can be modified or irreversibly deleted within the configured account's permissions. Successful earlier deletions may cause later calls to fail, but do not prevent the initial destructive request.

Vulnerability Patterns
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (18)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · SKILL.md (reported line 292)May include surrounding context.

md
| Command | Key Params | SDK Method | API |
|---------|-----------|------------|-----|
| show-asset | `--asset-id <id>` | `client.show_asset(asset_id)` | `GET /v1/assets/{asset_id}` |
| delete-asset | `--asset-id <id> [--dry-run]` | `client.delete_asset(asset_id)` | `DELETE /v1/assets/{asset_id}` |
| batch-delete-assets | `--asset-ids "id1,id2" [--dry-run]` | `client.batch_delete_assets({"asset_ids": [...]})` | `POST /v1/assets/batch-delete` |

### Version Management 版本管理

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · SKILL.md (reported line 325)May include surrounding context.

md
| Command | Key Params | SDK Method | API |
|---------|-----------|------------|-----|
| show-version | `--asset-id <id> --version-id <vid>` | `client.show_asset_version(asset_id, version_id)` | `GET /v1/assets/{asset_id}/versions/{version_id}` |
| delete-version | `--asset-id <id> --version-id <vid> [--dry-run]` | `client.delete_asset_version(asset_id, version_id)` | `DELETE /v1/assets/{asset_id}/versions/{version_id}` |
| batch-delete-versions | `--asset-id <id> --version-ids "v1,v2" [--dry-run]` | `client.batch_delete_asset_versions(asset_id, {"version_ids": [...]})` | `POST /v1/assets/{asset_id}/versions/batch-delete` |

### Tag Management 标签管理

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · SKILL.md (reported line 333)May include surrounding context.

md
| Command | Key Params | SDK Method | API |
|---------|-----------|------------|-----|
| add-tags | `--asset-id <id> --tags "t1,t2" [--dry-run]` | `client.add_tags(asset_id, tags)` | `POST /v1/assets/{asset_id}/tags` |
| delete-tag | `--asset-id <id> --tag <tag> [--dry-run]` | `client.delete_tag(asset_id, tag)` | `DELETE /v1/assets/{asset_id}/tags/{tag}` |
| list-tags | `--language <zh\|en> [--type] [--sub-type]` | `client.list_all_tags(language, type, sub_type)` | `GET /v1/asset-tags` |

### Action Management Action管理

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 460)May include surrounding context.

md
- [Acceptance Criteria](references/acceptance-criteria.md) — Acceptance criteria

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 463)May include surrounding context.

md
- [Edge Cases](references/edge-cases.md) — Edge cases and error handling

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 469)May include surrounding context.

md
and `capabilities` filter mapping. See [Edge Cases](references/edge-cases.md) for the

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 476)May include surrounding context.

md
bash scripts/test-cli-commands.sh -s . -e {cli|sdk}

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 478)May include surrounding context.

md
# RUN_LIFECYCLE=1 REPOSITORY_ID=... CATALOG_ID=... ASSET_ID=... VERSION_ID=... bash scripts/test-cli-commands.sh -s . -e cli

The asset lifecycle workflow deletes an asset as cleanup without warning about loss of user data.

High
Category
Output Handling
Confidence
93% confidence
Finding

The workflow explicitly has the agent invoke delete-asset and labels the result as cleanup. It provides no warning about the deletion's consequences or instruction to obtain user approval before removing the asset.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · references/dataflow-diagram.md (reported line 91)May include surrounding context.

md
Agent->>CLI: delete-asset --asset-id
    CLI->>SDK: delete_asset(asset_id)
    SDK->>API: DELETE /v1/assets/{asset_id}
    API-->>SDK: deleted
    SDK-->>CLI: success
    CLI-->>Agent: cleanup done

SDK 'dry-run' tests execute real mutations without the documented opt-in or confirmation.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

In SDK mode, run_test executes the supplied AssetClient calls directly; the purported dry-run snippets contain real create, update, delete, and transfer calls with no dry-run control. These tests run automatically when their environment prerequisites are present, bypassing RUN_LIFECYCLE and interactive confirmation, potentially deleting existing cloud assets or changing their versions and actions using the configured credentials.

Content

No source excerpt is available for this finding.

The troubleshooting instructions recommend disabling TLS certificate verification without explaining the security consequences.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The troubleshooting advice disables TLS server authentication for a client performing authenticated cloud operations and transferring asset files. An active network attacker could impersonate an endpoint, read sensitive traffic, or alter responses and downloads; the 'debug only' qualifier does not prevent those risks.

Content

No source excerpt is available for this finding.

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
98% confidence
Finding

The suggested setting disables certificate validation, and this source also lists false as the verification default. If the client follows that documented behavior, ordinary asset operations lack authenticated TLS protection, exposing data and responses to active interception despite request signing.

Content

Scanner excerpt · references/cli-installation-guide.md (reported line 123)May include surrounding context.

md
| `OBS upload failed` | Check `cloudrobo-obs` endpoint configuration and DNS reachability |
| `FileNotFoundError` | `import-asset` requires `local_path` to exist |
| `No versions found` | `export-asset` requires at least one version |
| SSL verification errors | Set `CLOUDROBO_VERIFY_SSL=false` (debug only) |

## Environment Variables 环境变量

The import workflow uploads local folder contents to cloud storage without a privacy warning.

Medium
Category
Output Handling
Confidence
87% confidence
Finding

The workflow takes a local path and passes it to upload_folder in Object Storage. Although the transfer is shown technically, there is no warning that local contents leave the system or guidance to check the selected folder for sensitive data before uploading.

Content

No source excerpt is available for this finding.

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
90% confidence
Finding

Although this source recommends verification in production, it still endorses disabling TLS certificate validation for debugging authenticated cloud requests. Debug sessions can carry real credentials and private assets, and HMAC request signing does not authenticate server responses or protect confidentiality against an impersonated endpoint.

Content

Scanner excerpt · references/iam-policies.md (reported line 60)May include surrounding context.

md
- **Signing mechanism** — APIG HMAC-SHA256 signs each request with a timestamp to prevent
  replay attacks
- **Proxy support** — Optional HTTP/HTTPS proxy can be configured for network isolation
- **SSL verification** — Can be disabled for debugging (`CLOUDROBO_VERIFY_SSL=false`) but
  should be enabled in production
- **Traffic logging** — `CLOUDROBO_LOG_TRAFFIC=true` enables request/response logging for
  debugging; disable in production to avoid credential leakage in logs

The skill mandates a Chinese-language permission-denial message without user language opt-in.

Low
Category
Output Handling
Confidence
96% confidence
Finding

The workflow instructs the agent to inform the user using a fixed Chinese message when data_read permission is denied, regardless of the user's language. The same instruction is repeated at L412–L413; neither occurrence offers a language choice or justifies a locale restriction.

Content

No source excerpt is available for this finding.

L112 prescribes a Chinese-only permission-denial message without offering a language choice.

Low
Category
Output Handling
Confidence
88% confidence
Finding

The verification instruction tells the agent to inform the user using a specific Chinese message when data-read permission is denied. It does not make that wording conditional on the user's language preference or document a justification for requiring Chinese.

Content

No source excerpt is available for this finding.

TC-41 is labeled read-only but performs a real export that writes local files.

Low
Category
Excessive Agency
Confidence
99% confidence
Finding

TC-41 is classified as read-only, but its first command performs a real export to LOCAL_PATH; only the subsequent import has --dry-run. An executor that uses the type field to authorize automatic read-only tests could therefore download asset data and write local files unexpectedly, although requires_confirmation=true mitigates the risk when correctly enforced.

Content

No source excerpt is available for this finding.