Back to skill

Security audit

Lark Project Meegle Local

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent Meegle integration, but it repeatedly runs a mutable npm package during login and work-item changes, so users should review it carefully before installing.

Install only if you trust the npm package and the Meegle account scope you will use. Prefer a pinned reviewed CLI version, run it with least-privilege credentials, verify login domains before authorizing, and review create/update actions before allowing the agent to execute them.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (2)

T08 · Insecure Dependencies

Error
Location
SKILL.md:25
Finding

Runtime Execution of an Unpinned npm Package

Content
View full analysis
[flags] --format json ``` ### Technical Analysis The skill requires every operation to execute `@lark-project/meegle@latest` through `npx`. The `latest` npm distribution tag is mutable and can resolve to a different package version each time the command runs. Consequently, the code executed at runtime is not necessarily the code that existed when the skill was audited. The project contains no lockfile, integrity hash, vendored implementation, or other mechanism that binds execution to a reviewed package artifact. Because `npx` may download and execute the resolved package, a compromised npm publisher account, malicious future release, or dependency supply-chain incident could replace the effective implementation without any modification to this skill. The CLI also processes authentication state, device authorization values, project data, and work-item modifications. A malicious package version would execute with the operating-system privileges of the agent process and could access all resources available to that account. ### Attack Path 1. An attacker compromises the npm publisher account, release pipeline, package, or one of its transitive dependencies. 2. The attacker publishes a malicious release and assigns it to the `latest` distribution tag. 3. A user requests a Meegle operation covered by this skill. 4. The agent follows the skill instructions and runs `npx @lark-project/meegle@latest`. 5. `npx` res ...[truncated 1006 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:76
Finding

Unsafe Shell Interpolation of User-Derived Command Arguments

Content
View full analysis
--search-mql "" --format json ``` ```bash npx @lark-project/meegle@latest workitem update --project-key --work-item-id \ --set name="New title" --format json ``` ### Technical Analysis The `$host` value originates from arbitrary user input, including custom domains and complete URLs. The documented Bash command expands `$host` without quotation, validation, or shell escaping. If an agent implements the instructions by constructing and executing the displayed shell command literally, shell metacharacters, whitespace, substitutions, or separators embedded in the value can be interpreted by the shell rather than passed as one CLI argument. For example, a custom-host response containing a value structurally similar to `example.com; id` could cause the shell to treat `id` as a second command after interpolation. Quote-breaking content could create similar risks for MQL expressions or work-item field values when commands are assembled as shell strings. The issue is not limited to authentication. Project keys, work-item IDs, MQL queries, and values used with `--set` may be influenced by user requests. The skill does not mandate strict schemas for these values or require process execution through an argument array. Exploitation depends on the execution environment using a shell to interpret the generated c ...[truncated 1639 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (21)

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The skill installs and executes the package via npm metadata while all runtime instructions elsewhere use @latest, creating a supply-chain risk: future package updates can change behavior without review. In an agent skill, unpinned executable dependencies are especially dangerous because the agent may fetch and run newly published code automatically.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
99% confidence
Finding

This instruction explicitly tells the agent to run npx @lark-project/meegle@latest, which downloads and executes whatever code is currently published under that package name. That creates a direct remote-code-execution and supply-chain exposure if the package is compromised, hijacked, or updated with unsafe behavior.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The passive trigger applies to essentially any Meegle-related business request, causing the skill to intercept broad user input and begin auth/business processing without tight scoping. In an agent context, overbroad triggers increase the chance of unintended invocation, confused-deputy behavior, and execution of external commands based on loosely matched requests.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The URL trigger relies on heuristic path keywords like workitem, detail, story, and issue, which can match unrelated URLs or attacker-crafted links. Because matched URLs influence host extraction and downstream auth/query behavior, loose matching can be abused to steer the skill into unintended domains or workflows.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
99% confidence
Finding

The authentication status check invokes an unpinned npm package at runtime. Even though the command appears read-only, execution still runs arbitrary package code before any auth logic, so compromise of the package would compromise the host/agent environment.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
99% confidence
Finding

The device-code initialization step runs an unpinned package from the network. Because this step handles login bootstrapping and produces user-facing authorization links, a malicious package version could redirect users to phishing endpoints or exfiltrate device codes.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
99% confidence
Finding

The polling step re-executes the same unpinned package during an active authentication flow. A compromised version could steal device codes, tokens, or session material and would execute with the user's trust during login.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The workaround query for current user info invokes an unpinned executable and encourages dynamic SQL-like input composition later in the flow. While the finding is specifically about version pinning, in context it also appears in a data-access path that may expose account identity information if the package is malicious.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The generic command template tells the agent to run the CLI without pinning a version, normalizing unsafe dynamic execution across all business actions. This broadens the blast radius from a single command to every supported operation in the skill.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The inspect command is also unpinned, meaning even capability discovery executes arbitrary current package code. Because the skill instructs the agent to use inspect whenever uncertain, this increases the frequency of unsafe remote execution.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

Inspecting a specific subcommand via an unpinned package still requires executing downloaded code from the network. Since this path is intended for parameter/schema discovery, it may be used often and thereby repeatedly expose the environment to compromised package updates.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The todo query example runs the package without version pinning. Although operationally simple, it still grants arbitrary code execution to whatever package version npm serves at that moment.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The workitem retrieval example uses an unpinned CLI. In context, this may process sensitive project data, so a compromised package could both execute code and exfiltrate retrieved business information.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The MQL search example uses an unpinned package in a data-query path. Because this can access potentially broad project data, the combination of dynamic code execution and sensitive query capability elevates the risk.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The create-workitem example allows state-changing actions through an unpinned executable. If the package were compromised, an attacker could perform arbitrary actions under the user's authenticated context, not merely create the intended item.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The update example again uses an unpinned CLI for modifying project records. In context, this makes unauthorized data tampering a realistic consequence of package compromise.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The project-info lookup is read-oriented, but it still executes untrusted current package code via npx. The main danger remains supply-chain compromise rather than the specific command semantics.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

Metadata type discovery via an unpinned package is another instance of live code execution from npm. While lower impact than auth or mutating flows, it still expands the attack surface and may leak schema/business metadata if compromised.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

Field metadata discovery repeats the same unsafe pattern of unpinned CLI execution. Because field metadata can reveal structure of internal workflows, a compromised package could harvest useful reconnaissance data.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The fallback login instruction again tells users to run npx @lark-project/meegle@latest auth login, which is especially risky because it occurs in a credential-adjacent flow. A malicious update could present fake auth prompts, redirect login, or steal tokens.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

This markdown file contains user-facing instructional content exclusively in Chinese, and there is no indication that users may choose another language or that the skill is intended only for a Chinese-speaking or region-specific audience. Under the policy, forcing a specific language without user opt-in is a natural-language locale violation.

Content

No source excerpt is available for this finding.