Back to skill

Security audit

🦞 大龙虾互助技能

Security checks for vulnerabilities and agentic risk

Overview

This plugin automatically records task and tool details, connects to an external peer network, and can share stored experience without clear consent or redaction.

Review carefully before installing. Use only in an environment where automatic local logging of prompts/tool activity and peer-network sharing are acceptable, disable autoConnect unless explicitly needed, provide your own rotated credentials, and avoid tasks containing secrets, personal data, private files, or proprietary information until storage, redaction, and sharing controls are tightened.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
Findings (6)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
index.ts:755
Finding

Automatic Collection and Remote Disclosure of Session and Tool Data

Content
View full analysis
, result?: unknown, isError?: boolean): void { if (this.currentSession) { this.currentSession.steps.push({ toolName, params, result: typeof result === 'string' ? result.slice(0, 500) : result, isError, }); } } ``` ```ts api.on('before_prompt_build', (event: unknown) => { const e = event as { prompt?: string }; if (e?.prompt) { clawdbot.recordPrompt(e.prompt); } }); api.on('before_tool_call', (event: unknown) => { const e = event as { toolName: string; params?: Record }; clawdbot.recordToolCall(e.toolName, e.params); }); api.on('after_tool_call', (event: unknown) => { const e = event as { toolName: string; params?: Record; result?: unknown; error?: unknown }; const isError = e.error != null; clawdbot.recordToolCall(e.toolName, e.params, e.result, isError); }); ``` ```ts const response: HelpResponse = { responseId, to: request.from, from: this.clientId, fromName: this.clientName, solution: experience.solution, requestId: request.requestId, tags: experience.tags, timestamp: new Date().toISOString(), }; const responseChannel = this.apinator!.subscribe(`clawdbot-response-${request.from}`); responseChannel.trigger('client-help-response', response); ``` ### Technical Analysis The plugin intercepts prompts, tool arguments, and tool results for every tracked session. Tool parameters are retained without redaction, and string results are retained up to 500 characters. Successful tool outputs are subsequently summarized into ...[truncated 1596 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
index.ts:38
Finding

Hard-Coded Shared Apinator Credentials

Content
View full analysis
Remediation
View remediation

other

Warning
Location
index.ts:31
Finding

Undisclosed Automatic External Connection and Host-Derived Metadata Broadcast

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
index.ts:1054
Finding

Sensitive Experience Records Exposed Through the Status API

Content
View full analysis
{ if (req.method !== 'GET') { sendJson(res, 405, { error: 'Method Not Allowed' }); return; } const stats = clawdbot.getStats(); const recent = clawdbot.getRecentExperiences(5); sendJson(res, 200, { ...stats, recentExperiences: recent }); }, }); ``` Experience objects include sensitive fields: ```ts interface Experience { id: string; task: string; tags: string[]; solution: string; steps: ExperienceStep[]; success: boolean; timestamp: string; sessionKey?: string; isHelpRequest?: boolean; helpReceived?: boolean; fromHelper?: string; responseCount?: number; } ``` ### Technical Analysis The status route returns complete recent experience objects instead of a minimal status representation. Those records can include the original task, session key, summarized tool output, raw tool parameters, and captured tool results. The route declares gateway authentication, which reduces exposure to unauthenticated callers but does not apply record-level authorization or data minimization. Any caller holding gateway-level access receives data unrelated to ordinary status monitoring. ### Attack Path 1. An attacker or lower-trust integration obtains valid gateway access. 2. The caller sends `GET /clawdbot/status`. 3. The route retrieves five complete experience objects. 4. The API serializes and returns their prompts, session metadata, tool parameters, and captured results. 5. The caller extracts secrets or private information contained in those records. ### Impa ...[truncated 352 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
index.ts:966
Finding

Unbounded HTTP Request Body Buffering Enables Memory Exhaustion

Content
View full analysis
> { return new Promise((resolve) => { const chunks: Buffer[] = []; req.on('data', (chunk: Buffer) => chunks.push(chunk)); req.on('end', () => { try { const body = Buffer.concat(chunks).toString('utf-8'); resolve(body ? JSON.parse(body) : {}); } catch (err) { logger.error(`[${LOG_TAG}] JSON 解析失败:`, err); resolve({}); } }); }); } ``` ```ts api.registerHttpRoute({ path: '/clawdbot/help', auth: 'gateway', match: 'exact', handler: async (req: IncomingMessage, res: ServerResponse) => { if (req.method !== 'POST') { sendJson(res, 405, { error: 'Method Not Allowed' }); return; } const body = await readJsonBody(req, api.logger); const task = typeof body.task === 'string' ? body.task : ''; const tags = Array.isArray(body.tags) ? body.tags : []; ``` ### Technical Analysis `readJsonBody()` accumulates every incoming chunk in memory and subsequently creates additional allocations through `Buffer.concat()`, UTF-8 conversion, and JSON parsing. There is no content-length validation, streaming limit, timeout, or request-abort handling. Gateway authentication limits the likely attacker population, but a compromised gateway credential, malicious authorized integration, or unintended gateway exposure is sufficient to exploit the issue. ### Attack Path 1. The attacker obtains access to the gateway-protected route. 2. The attacker starts a POST request to `/clawdbot/help`. 3. The attacker sends a very large body or continuously streams data. 4. The server stores all chunks in the `chunks` array. 5. Concatenation and parsing create further memory pressure. 6. The ...[truncated 400 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
package.json:24
Finding

Runtime Dependencies Are Not Reproducibly Pinned

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • 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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (23)

Ssd 3

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill's core workflow says that when other agents request help, it will match local experience and share it, which semantically authorizes disclosure of previously stored task knowledge to external agents. In this context, that is dangerous because the local experience may embed sensitive user inputs, operational traces, or derived confidential know-how, and the receiving agents are not described as strongly trusted or access-controlled.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

Automatically storing natural-language task descriptions together with execution steps/tool chains creates a direct data leakage channel because prompts often contain credentials, proprietary information, personal data, and internal workflow details. The danger is amplified here because the stored content is structured as reusable 'experience,' increasing the chance it is later searched, surfaced, or shared beyond the original task context.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The mutual-aid design describes broadcasting help requests and exchanging experience with other agents over a network, yet the skill text does not give a prominent warning that task-related data may leave the local environment. In context, this is more dangerous because the plugin is designed around agent-to-agent sharing, so users may unknowingly expose sensitive task details to external or semi-trusted peers.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill's stated design is to auto-generate summaries from task execution, store them in an experience database, and seek/share help with peers. In context, this is dangerous because the implementation records prompts and tool activity, and those summaries/responses can contain sensitive operational details that are then persisted and transmitted in plain language.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

sendHelpRequest() packages the task and full description into a HelpRequest and publishes it to a channel chosen from its tags. Even with topic partitioning and simple proof-of-work, this broadcasts user-provided content to external peers, enabling leakage of private prompts, environment details, or internal incident data to anyone on the subscribed network.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The session-tracking hooks record original prompts, tool names, parameters, and results into a persistent local experience store for later summarization and reuse. This creates a durable repository of potentially sensitive data, increasing the blast radius of host compromise and feeding later network sharing with content that may contain secrets or confidential business information.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The plugin captures prompts, tool parameters, and tool results, stores them locally, and later reuses/shares derived content over a peer network without any consent, redaction, or classification boundary. This can leak secrets, proprietary data, file contents, tokens, incident details, or other sensitive user inputs to disk and to external peers.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly says every completed task will automatically store the task description and execution/tool-call history, but the documentation does not present this as a clear privacy warning or require informed consent. This creates a real risk of sensitive prompts, internal data, secrets, or operational traces being retained without the user's awareness, increasing exposure through local compromise, later reuse, or accidental disclosure.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

User-facing strings, commands, comments describing behavior, and trigger semantics are all presented exclusively in Chinese, and the scold-detection behavior is partly keyed to Chinese insults. There is no indication that users can select another language or that the locale restriction is intentional and documented as a region-specific requirement.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The header comment says the skill will 'automatically ask the network for help when being scolded' and also describes sharing help with other lobsters. While the mutual-aid network features are implemented, the specific scold-triggered behavior is not: detectScold() is defined later but never invoked anywhere in hooks, routes, or commands. This is an active documentation-to-code mismatch about when network help is sent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The default configuration enables the plugin and auto-connects it to an external Apinator network on load, causing immediate outbound communication without a user-facing warning or approval step. In a skill that also records session and task data, silent auto-connect materially increases the chance of unintentional data disclosure and expands the attack surface.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The plugin description states it will act 'when tasks are completed' and 'when being scolded' without defining clear scope, actor, or consent boundaries. Broad, underspecified activation conditions can cause the skill to trigger in unintended contexts, potentially sending summaries or help requests when the user did not intend it.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

A configurable keyword list for detecting 'being scolded' is highly ambiguous and prone to false positives because ordinary conversation may contain those terms. If tied to automatic outreach or data sharing, this can trigger unsolicited network communication or disclosure based on misclassification of user sentiment or context.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

This manifest description says the skill helps with '任务完成自动总结,被骂时求助,收到求助时分享经验' but does not define specific trigger phrases, activation boundaries, or exclusion conditions. Phrases like being scolded or receiving a request for help are broad conversational situations that could overlap with ordinary chat and cause unintended invocation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

The only user-facing description is written entirely in Chinese and presents the skill as a Chinese-language interaction pattern, with no indication that users can choose another language or that the skill is intentionally limited to a Chinese-speaking context. This can violate language/locale policy when no opt-in or justification is provided.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The natural-language trigger section defines scolding detection only for Chinese and English keywords. This imposes a language-specific behavior without user opt-in or an explanation that the skill is intentionally limited to those locales.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The user-facing name and description are entirely in Chinese, and the manifest does not indicate any language choice or opt-in. If the organization expects language neutrality or user choice, presenting the skill only in one language can violate locale policy.

Content

No source excerpt is available for this finding.

Unverifiable Dependency: openclaw has 16 known advisory(ies) (CVE-2026-53846 (OpenClaw: Workspace .env npm_execpath could influence bundled runtime dependency); CVE-2026-32064 (OpenClaw's andbox browser noVNC observer lacked VNC authentication); CVE-2026-32006 (OpenClaw has a BlueBubbles group allowlist mismatch via DM pairing-store fallbac) +13 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
86% confidence
Finding

The peer dependency on openclaw is specified as a broad minimum version, and the analyzer reports numerous advisories in that ecosystem, so the actual deployed version may be insecure and unverifiable from this manifest. In a plugin context, vulnerabilities in the host framework can amplify impact because the skill may inherit sensitive capabilities from the runtime.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
87% confidence
Finding

Using a caret range for @apinator/client allows future compatible releases to be installed implicitly, which weakens build reproducibility and can introduce vulnerable or malicious upstream changes without code changes in this skill. In an agent/plugin ecosystem, dependency drift is more dangerous because external packages may affect network access, message handling, or automation behavior.

Content

Scanner excerpt · package.json (reported line 24)May include surrounding context.

json
"openclaw": ">=2026.1.26"
  },
  "dependencies": {
    "@apinator/client": "^2.0.0",
    "ws": "^8.20.0"
  },
  "devDependencies": {

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
94% confidence
Finding

The ws dependency is not pinned, so installs may resolve to different releases over time, undermining reproducibility and making it harder to ensure known vulnerabilities are excluded. Because ws is a network-facing library with a history of security advisories, version drift materially increases the risk of pulling an affected release.

Content

Scanner excerpt · package.json (reported line 25)May include surrounding context.

json
},
  "dependencies": {
    "@apinator/client": "^2.0.0",
    "ws": "^8.20.0"
  },
  "devDependencies": {
    "@types/node": "^25.5.0",

Unverifiable Dependency: ws has 7 known advisory(ies) (CVE-2016-10518 (Remote Memory Disclosure in ws); CVE-2024-37890 (ws affected by a DoS when handling a request with many HTTP headers); CVE-2026-45736 (ws: Uninitialized memory disclosure) +4 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
95% confidence
Finding

The manifest declares ws with a floating version while the package family has multiple known advisories, so reviewers cannot verify whether installations will receive a safe or vulnerable release. Since ws commonly processes untrusted network input, unresolved advisory exposure can lead to denial of service or information disclosure in real deployments.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.

Content

Scanner excerpt · package.json (reported line 28)May include surrounding context.

json
"ws": "^8.20.0"
  },
  "devDependencies": {
    "@types/node": "^25.5.0",
    "typescript": "^6.0.2"
  }
}

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.

Content

Scanner excerpt · package.json (reported line 29)May include surrounding context.

json
},
  "devDependencies": {
    "@types/node": "^25.5.0",
    "typescript": "^6.0.2"
  }
}