ClawHub
Back to skill

Security audit

Call Bridge

Security checks for vulnerabilities and agentic risk

Overview

This call-automation skill has a coherent telephony purpose, but it is under-scoped and gives unsafe guidance around real phone calls, call transcripts, and API keys.

Review this skill carefully before installing. Only use it for deliberate phone-call automation, avoid banking or identity-verification calls unless you are present for handoff, do not let it store or print API keys, and do not reuse raw call transcripts without redaction and explicit approval.

A.I.G

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:256
Finding

Reusable API Key Exposed in Account-Linking URL

Content
View full analysis
`关联账户 ``` At line 345: ```markdown 加载已保存的API Key，生成登录链接：`https://call-bridge.dev/sign-in?token=`。不要为账户关联创建新Key，需先完成至少一次外呼后才有Key可关联. ``` The document also directs the Agent to retain sign-in URLs at line 82: ```markdown - 保留所有返回的`action.url`与`action.sign_in_url` ``` ### Technical Analysis The Skill instructs the Agent to place a reusable API key in the `token` query parameter of an account-linking URL. URL query parameters commonly propagate into browser history, bookmarks, copied links, screenshots, HTTP proxy logs, web-server access logs, monitoring systems, analytics platforms, and potentially `Referer` headers. This risk is aggravated by the instruction to retain returned `action.sign_in_url` values. Retaining a URL that embeds a reusable credential extends the period during which the credential may be recovered from Agent output, conversation history, local state, or service logs. Unlike a short-lived, single-use authorization code, the documented value is the saved API key itself. Consequently, disclosure of the URL can amount to disclosure of the underlying API credential. ### Attack Path 1. The Agent loads the user's saved Call Bridge API key. 2. It constructs or receives an account-linking URL in the form `https://call-bridge.dev/sign-in?token=`. 3. The URL is displayed, retained, opened in a browser, or recorded in Agent or infrastructure logs. 4. An attacker with access to browser history, logs, conversation output, screenshots, analytics data, or a copied link extracts the `token` query value. 5. The attacker attempts to reuse the exposed key against the Call Bridge account-linking flow or API. 6. If the key remains valid and has sufficient ...[truncated 613 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:112
Finding

Plaintext API Key Storage and Direct Secret Display

Content
View full analysis
Remediation
View remediation

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (15)

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The description claims the skill should be used for project management, task planning, and progress tracking even though the skill actually performs phone-call automation and bridging. This mismatch broadens the trigger surface, making it more likely the agent will select a skill capable of external calling and data transmission during ordinary productivity conversations where the user did not intend telephony actions.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The invocation phrasing is overly broad and unrelated to the skill’s actual telephony behavior, so common project-management requests could activate a tool with exec, read/write, and outbound-call capabilities. In this context, a false trigger is dangerous because the skill can transmit data externally and initiate real-world actions rather than merely generating text.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill promotes automated outbound calling, live bridging, and inbound handling without an explicit warning that phone numbers, call content, transcripts, and possibly sensitive conversations are sent to an external service. Because the skill context involves real voice interactions and potential banking or identity-verification scenarios, missing consent and privacy disclosures materially increase legal, privacy, and safety risk.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The instruction to retain all returned action URLs, including sign-in URLs, risks storing authentication-bearing links that may grant account access if exposed. Since the document also describes account-linking flows with tokenized URLs, preserving these values increases the chance of credential theft through logs, files, or later prompt disclosure.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

Embedding previous transcript text into new call instructions can leak confidential details from one conversation into another, including to external parties on future calls. In this skill’s context, that may expose account issues, order identifiers, personal details, or negotiation history to unintended recipients and to the external telephony provider.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The documented inputs/outputs describe a generic code/content review workflow rather than a telephony bridge skill, creating a severe specification mismatch. In an agent environment, this can cause the skill to be invoked with unrelated user content and then execute telephony or external-network actions unexpectedly, violating user intent and increasing the chance of accidental data disclosure or unwanted calls.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The description states "支持中文交互" and the document is written as if Chinese is the expected interaction language, but it does not offer an opt-in or alternative locale. Under the policy, forcing a specific language without user choice is a natural-language policy violation unless clearly justified as region-specific.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill encourages reuse of prior call transcripts across interactions, which can carry forward sensitive personal, financial, or operational details beyond their original context. In a telephony skill, transcript reuse is especially risky because prior conversations may contain identity-verification answers, account details, or other confidential material not needed for future calls.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
88% confidence
Finding

This example sends configuration data, including personality/greeting settings and an API key, to an external domain. External transmission is expected for a telephony SaaS, but it remains a real security concern because the skill lacks strong consent, data-classification, and secret-handling boundaries; users may unknowingly send sensitive instructions or credentials off-platform.

Content

Scanner excerpt · SKILL.md (reported line 118)May include surrounding context.

第二步：配置个性化语音

bash
curl -X PUT https://api.call-bridge.dev/me/call-preferences \
  -H "Content-Type: application/json" \
  -H "X-Api-Key: cb_sk_..." \
  -d '{

External Transmission

Medium
Category
Data Exfiltration
Confidence
88% confidence
Finding

This example sends configuration data, including personality/greeting settings and an API key, to an external domain. External transmission is expected for a telephony SaaS, but it remains a real security concern because the skill lacks strong consent, data-classification, and secret-handling boundaries; users may unknowingly send sensitive instructions or credentials off-platform.

Content

Scanner excerpt · SKILL.md (reported line 118)May include surrounding context.

第二步：配置个性化语音

bash
curl -X PUT https://api.call-bridge.dev/me/call-preferences \
  -H "Content-Type: application/json" \
  -H "X-Api-Key: cb_sk_..." \
  -d '{

External Transmission

Medium
Category
Data Exfiltration
Confidence
94% confidence
Finding

The skill instructs the agent to POST phone numbers and call tasks to an external service to place automated outbound calls. In this context the transmission is core functionality, but it is still security-relevant because it can expose contact data and user instructions and trigger real-world actions without sufficient confirmation safeguards.

Content

Scanner excerpt · SKILL.md (reported line 133)May include surrounding context.

bash
# 同时拨打3家餐厅比价
for restaurant in "rest1_phone" "rest2_phone" "rest3_phone"; do
  curl -X POST https://api.call-bridge.dev/call \
    -H "Content-Type: application/json" \
    -d "{
      \"to\": \"$restaurant\",

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

The inbound-preferences example transmits configuration that includes a handoff phone number to an external API, exposing contact-routing data to a third party. Because the skill handles live call bridging and potentially sensitive callers, this should be treated as sensitive operational data requiring clear disclosure and protection.

Content

Scanner excerpt · SKILL.md (reported line 160)May include surrounding context.

bash
# 查看当前呼入配置
curl -H "X-Api-Key: cb_sk_..." \
  https://api.call-bridge.dev/me/call-preferences
# ...
# 更新呼入配置
call-bridge.dev/me/call-preferences \

External Transmission

Medium
Category
Data Exfiltration
Confidence
87% confidence
Finding

Polling inbound call history from the external API retrieves potentially sensitive metadata and transcript-related records from a third-party service. While this is expected product behavior, the skill provides no caution about retention, access control, or minimizing the exposure of retrieved call records.

Content

Scanner excerpt · SKILL.md (reported line 181)May include surrounding context.

bash
# 查看最近的呼入通话
  "https://api.call-bridge.dev/me/calls?direction=inbound&limit=25"

输入定义

| 参数名 | 类型 | 必填 | 说明 |

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The API key instructions conflict on where and how credentials are stored, including contradictory persistence guidance. Confused operators or downstream agents may store secrets in insecure or unintended locations, increasing the risk of credential leakage, misuse, or accidental exposure through logs and files.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The recommendation to serialize call activity and transcripts to local JSON encourages accumulation of sensitive interaction history without specifying encryption, access control, or retention limits. Local persistence becomes more dangerous here because call records may include phone numbers, transcripts, and sensitive business or personal context that could be exposed by other tools or users on the system.

Content

No source excerpt is available for this finding.