Back to skill

Security audit

地鼠AI剪辑标准 Skill

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent AICut video-editing assistant, but it exposes broad local-media and generic desktop-bridge authority that users should review before installation.

Install only if you trust the AICut desktop app and want an agent to control editing projects and local media. Authorize only narrow project folders, avoid Desktop or home-directory roots, require header-based tokens for local media, do not use query-string tokens, confirm any cloud upload or tunnel/public-frame sharing, and revoke media roots when finished.

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

Warning
Location
reference.md:371
Finding

Optional Authentication and Query-String Credentials for Local Media Access

Content
View full analysis
Remediation
View remediation
` or the existing `x-aicut-local-media-token` header. 3. Generate high-entropy, short-lived, per-session tokens rather than relying on static shared secrets. 4. Bind the media service to `127.0.0.1` or an equivalent loopback interface by default. Require an explicit security warning and hardened configuration before allowing non-loopback binding. 5. Enforce a strict allowlist of trusted origins and reject cross-origin browser requests by default. Do not combine permissive CORS settings with local-file access. 6. Apply `Referrer-Policy: no-referrer`, `Cache-Control: no-store`, and appropriate response headers to media and authentication-related responses. 7. Redact authorization headers, tokens, and media paths from application logs, proxy logs, telemetry, and error reports. 8. Keep the existing realpath and allowed-root validation, and ensure authorization is performed after canonicalization but before opening or streaming the file. 9. Prefer per-file opaque handles created through explicit user selection instead of exposing arbitrary path-based retrieval within a broad root. 10. Add automated security tests covering missing credentials, invalid and expired tokens, query-string token rejection, cross-origin requests, encoded traversal attempts, symlink escapes, and unauthorized Range requests. ]]>
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 642)May include surrounding context.

md
如果只能读取一个文件,优先读取本 `SKILL.md`,不要把 JSON 模板当成另一套 Skill。

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The metadata and skill framing are fixed to a Chinese-language, China-market editing workflow without any opt-in or locale negotiation. While not directly enabling code execution, it can cause the agent to ignore user language preferences or regional compliance expectations, which increases the chance of unsafe misunderstandings around exports, privacy review, or platform-specific publishing assumptions.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger list is broad enough to hijack many ordinary user requests about editing, media analysis, or exports and force the agent into this skill's specialized workflow. In this skill's context, that is risky because the workflow directs the agent to interact with local desktop services, import local media, and potentially modify projects, so accidental activation can expand the skill's authority beyond what the user explicitly intended.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

SQP-3 applies to all file types and covers language/locale policy violations. This markdown file presents all instructions, examples, and operational guidance exclusively in Chinese, with no opt-in mechanism or note that the skill is intentionally region-specific.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The documentation explicitly exposes a generic bridge_call tool that can invoke arbitrary window.aicutAI methods rather than a constrained, task-specific subset. In an agent-facing skill, this becomes a powerful ambient capability that can be abused to access local-media functions, enumerate authorized roots, read files, manipulate projects, or invoke future methods added to the bridge, creating a broad attack surface well beyond the stated editing-assistant role.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The title and throughout the document the skill is presented as operating in Chinese, and there is no indication that users may choose another language or locale. Under the language/locale policy rule, forcing a specific language without opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.