Back to skill

Security audit

Pikpak

Security checks for vulnerabilities and agentic risk

Overview

This skill is for managing a PikPak cloud drive, but it asks users to build and run an unpinned third-party command-line tool that will receive account credentials and can delete cloud files.

Install only if you are comfortable giving a locally built third-party CLI access to your PikPak account. Prefer a pinned, reviewed release with checksum or signature verification, and require explicit confirmation before any delete or bulk-change operation.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T03 · Remote Payload Retrieval and Execution

Error
Location
SKILL.md:23
Finding

Mutable Remote Source Is Downloaded, Compiled, and Executed

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:23-37
Additional Location: README.md:4
Vulnerability Type: Remote payload retrieval and execution through an unpinned third-party repository
Risk Level: High

Vulnerable Code

bash
git clone https://github.com/52funny/pikpakcli
bash
go build
text
./pikpakcli

The project also references another mutable repository location:

text
Install skill https://github.com/CN-CODEGOD/PIKSKILL/blob/main/SKILL.md

Technical Analysis

The Skill instructs the Agent to clone the current state of a remote GitHub repository, compile it, and execute the resulting binary. The source is not pinned to a reviewed commit hash or immutable release, and the instructions do not require checksum or signature verification.

Consequently, the code that ultimately executes can change after this Skill has been reviewed. The audited project does not include the downloaded source, so the behavior of the resulting executable cannot be established from the package itself. The README.md installation link similarly references the mutable main branch.

This behavior is best classified as remote payload retrieval and execution because a remote party controlling or compromising the referenced repository can modify the effective payload without changing the audited Skill package.

Attack Path

  1. An attacker compromises the referenced repository, a maintainer account, or the upstream publishing process.
  2. The attacker modifies the repository's default branch to include malicious Go code.
  3. The Agent follows SKILL.md and clones the mutable repository without selecting a trusted commit.
  4. The Agent runs go build, incorporating the attacker's code into the local executable.
  5. The Agent executes ./pikpakcli under the current user's identity.
  6. During setup, the user supplies PikPak credentials as required by the Skill.
  7. T ...[truncated 1015 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin the repository to a specific, reviewed commit hash rather than cloning and building the default branch.
  2. Prefer a signed, versioned release obtained from a trusted release channel.
  3. Publish and verify a cryptographic checksum, such as SHA-256, before executing any downloaded artifact.
  4. Verify release signatures against a documented and trusted maintainer key.
  5. Vendor the reviewed source into the Skill package where feasible so the audited and executed code are identical.
  6. Pin the README.md installation reference to an immutable commit URL instead of the mutable main branch.
  7. Review the complete dependency graph and use reproducible builds before distributing binaries.
  8. Run the utility with least privilege in a restricted environment that does not expose unrelated files, credentials, or environment variables.
  9. Avoid passing credentials to an unverified executable. Use a narrowly scoped authentication mechanism where supported and protect configuration files with restrictive filesystem permissions.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Hidden Instructions

High
Category
Prompt Injection
Confidence
60% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · README.md (reported line 1)May include surrounding context.

md
# pikpakcLi-skill
 # 安装

安装skill https://github.com/CN-CODEGOD/PIKSKILL/blob/main/SKILL.md

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
93% confidence
Finding

The skill includes a direct destructive command example that deletes a path in cloud storage without any surrounding safety checks, confirmation flow, or path validation guidance. In an agent-operated setting, this is dangerous because a model may generalize from the example and execute deletion on user data based on incomplete or mistaken prompts, causing irreversible loss.

Content

Scanner excerpt · SKILL.md (reported line 101)May include surrounding context.

  • 按完整路径删除文件

    bash
    pikpakcli delete /Movies/Peppa_Pig.mp4
    
  • 使用 -p 指定父目录后删除其中的文件或文件夹

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly documents file and folder deletion commands but provides no warning about permanence, confirmation expectations, or the need to verify target paths before execution. In an agent context, this increases the chance of accidental destructive actions against user cloud storage, especially if commands are composed from ambiguous prompts or inferred paths.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The document switches from English to Chinese for operational instructions, which can impose a language requirement on users without opt-in or explanation. This creates a natural-language policy concern because the skill does not state that it is intended for Chinese-speaking users or provide an alternative language option.

Content

No source excerpt is available for this finding.