Back to skill

Security audit

lovart-api

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence, citing broad triggers and workflow concerns; the scan was partial, so the evidence is incomplete.

Treat the suspicious verdict as a reason for caution. Before installing, consider whether you are comfortable with the reported remote project changes and external media uploads, and review the upload destination and intended permissions. The scan covered 50% of components and one analyzer was degraded, so its evidence is incomplete.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (10)

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Content

Scanner excerpt · SKILL.md (reported line 9)May include surrounding context.

md
creation request in any language — draw, generate, create, design, make, 画, 生成,
  制作, 创作, 设计 combined with image, video, audio, music, song, BGM, poster, etc.
  (2) Lovart project/thread management — 项目, 对话, project, thread, conversation,
  history, 历史, 切换, switch. You CAN generate directly - never say you cannot.
user-invocable: true
version: 1.1.1
author: Lovart (lovartai)

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, file_write, network.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Project and history triggers are not restricted to Lovart.

Medium
Category
Trigger Abuse
Confidence
97% confidence
Finding

The mandatory routing rule applies to generic projects, history, conversations, and settings without requiring Lovart context. This can redirect unrelated requests into Lovart operations, exposing unrelated project metadata or changing the wrong application's state rather than honoring the user's intended scope.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 78)May include surrounding context.

md
- `"done"` — Generation complete. Send the downloaded files to the user.
- `"pending_confirmation"` — A high-cost tool (e.g. video, or a premium-quality image variant) needs user approval before credits are consumed.
  **You MUST ask the user for explicit confirmation before proceeding. Do NOT auto-confirm.**
  1. Show the user: "This will cost approximately {estimated_cost} credits. Shall I proceed? (yes/no)"
  2. **WAIT for user response.** Only if user explicitly says yes/confirm/proceed, run:
     `confirm --thread-id THREAD_ID --json --download`

Troubleshooting guidance steers the assistant to alternate models after content-moderation refusals.

Medium
Category
Prompt Injection
Confidence
95% confidence
Finding

The troubleshooting guidance identifies content-moderation refusal as a failure cause and recommends switching models or simplifying policy-sensitive prompts. Without distinguishing safety refusals from technical failures, this encourages routing rejected requests to providers with different safeguards.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 134)May include surrounding context.

md
**Step 1: `config --json`**
- Check local state (`~/.lovart/state.json`) for `active_project`
- If `active_project` is set → proceed to Step 2. Do NOT create a new project. Do NOT ask the user.
- If `active_project` is missing → ask the user: "Do you have an existing Lovart project ID, or should I create a new one?" **WAIT for their answer.**
- Save with: `project-add --project-id PID --name "name"`

The attachment workflow uploads user media without a privacy warning.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The workflow directs the assistant to upload user-provided media to an external CDN without first establishing that the user authorized third-party processing. Uploading references is legitimate for this service, but the skill's broad automatic activation makes unexpected transfer of sensitive media more plausible.

Content

No source excerpt is available for this finding.

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · scripts/agent_skill.py (reported line 419)May include surrounding context.

python
def chat(self, prompt: str, project_id: Optional[str] = None,
             attachments=None, thread_id=None,
             timeout=None, auto_create_project=True,
             prefer_models=None, include_tools=None, exclude_tools=None,
             mode: Optional[str] = None, subjects=None, kits=None) -> dict:
        """Send → poll → get result. One-shot convenience method.

Commands documented as local-only can modify remote Lovart projects.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The ostensibly local command block performs authenticated remote renames during project-add and project-rename when credentials are available. A user merely registering a local project with a friendly name can therefore change the remote project's name, while swallowed exceptions conceal whether synchronization succeeded.

Content

No source excerpt is available for this finding.

Cloud Storage Exfiltration

Medium
Category
Data Exfiltration
Confidence
55% confidence
Finding

Data is uploaded to cloud storage (S3 / GCS / Azure Blob). This may be a legitimate backup or exfiltration to an external bucket. Manual review is recommended.

Content

Scanner excerpt · scripts/agent_skill.py (reported line 1065)May include surrounding context.

python
print(json.dumps({"project_id": pid}))

        elif args.command == "upload":
            url = skill.upload_file(args.file)
            print(json.dumps({"url": url}))

        elif args.command == "upload-artifact":