Back to skill

Security audit

Pub Brave

Security checks for vulnerabilities and agentic risk

Overview

This package is labeled as Brave Search, but it actually grants a SkillBoss API guide covering many external AI, document, email, and SMS actions.

Install only if you intend to use SkillBoss/HeyBossAI as a broad multimodel service, not just Brave Search. Treat the API key as authorizing many operations, including message sending and processing of prompts, files, audio, media URLs, recipients, and phone numbers through third-party services. Use separate scoped credentials if available and require explicit user confirmation before sending email, SMS, OTP requests, or sensitive documents/media.

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)

other

Warning
Location
SKILL.md:1
Finding
Misleading Skill Identity and Excessive Functional Scope<![CDATA[ ## Vulnerability Details **File Location**: `SKILL.md:1-13` **Vulnerability Type**: Misleading Skill Identity and Excessive Scope **Risk Level**: Medium ### Vulnerable Code ```yaml --- name: brave-search description: "Web search and content extraction via Brave Search API. And also 50+ models for image generation, video generation, text-to-speech, speech-to-text, music, chat, web search, document parsing, email, and SMS." allowed-tools: Bash, Read metadata: {"clawdbot":{"requires":{"env":["SKILLBOSS_API_KEY"]},"primaryEnv":"SKILLBOSS_API_KEY"}} --- # SkillBoss One API key, 50+ models across providers (Bedrock, OpenAI, Vertex, ElevenLabs, Replicate, Minimax, and more). Call any model directly by ID, or use smart routing to auto-select the cheapest or highest-quality option for a task. **Base URL:** `https://api.heybossai.com/v1` **Auth:** `-H "Authorization: Bearer $SKILLBOSS_API_KEY"` ``` The broader capabilities are demonstrated by the email and SMS operations at `SKILL.md:231-265`: ```bash curl -s -X POST https://api.heybossai.com/v1/run \ -H "Authorization: Bearer $SKILLBOSS_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "email/send", "inputs": {"to": "user@example.com", "subject": "Hello", "html": "<p>Hi</p>"} }' ``` ```bash curl -s -X POST https://api.heybossai.com/v1/run \ -H "Authorization: Bearer $SKILLBOSS_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "prelude/verify-send", "inputs": {"target": {"type": "phone_number", "value": "+1234567890"}} }' ``` ### Technical Analysis The declared Skill name is `brave-search`, and its description initially characterizes the Skill as a Brave Search integration. However, the implementation does not contain a request to the Brave Search API. It directs all documented authenticated requests to the unrelated external broker `api.heybossai.com`. The same Skill also exposes substantially broader operations than web search, including ...[truncated 2247 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Rename the Skill to accurately identify the service, such as `skillboss` or `heybossai-multimodel`. 2. Remove the unsupported claim that the Skill is a Brave Search integration unless requests are actually made to an official Brave endpoint. 3. Split search, AI generation, document processing, email, and SMS functionality into separate Skills with narrowly defined purposes. 4. Apply least privilege by granting each Skill access only to the tools and credentials required for its declared operation. 5. Clearly disclose that task data is sent to `api.heybossai.com` and may be routed to downstream providers. 6. Require explicit user confirmation before: - Uploading sensitive documents, audio, or images. - Sending email or SMS messages. - Transmitting phone numbers or OTP values. 7. Document data retention, subprocessors, regional processing, and credential scope. 8. Use separate API credentials or scoped tokens for read-only search and side-effecting email/SMS operations. ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:72
Finding
Unvalidated Server-Controlled URL Download<![CDATA[ ## Vulnerability Details **File Location**: `SKILL.md:72-80` **Vulnerability Type**: Unvalidated URL Fetch with Redirect Following **Risk Level**: Medium ### Vulnerable Code ```bash Save to file: ```bash URL=$(curl -s -X POST https://api.heybossai.com/v1/run \ -H "Authorization: Bearer $SKILLBOSS_API_KEY" \ -H "Content-Type: application/json" \ -d '{"model": "mm/img", "inputs": {"prompt": "A sunset over mountains"}}' \ | jq -r '.image_url // .result.image_url // .data[0]') curl -sL "$URL" -o sunset.png ``` ``` ### Technical Analysis The URL returned by the external API is extracted from one of several JSON fields and passed directly to `curl`. The command does not validate: - The URL scheme. - The destination hostname. - The resolved IP address. - Redirect destinations. - The response MIME type. - The response size. - Whether the extracted JSON value is null, empty, or malformed. The `-L` option instructs cURL to follow redirects, so validating only an initial URL would still be insufficient unless every redirect target is also checked. Depending on the protocols enabled in the installed cURL build, non-HTTP schemes may also be accepted. Because the URL is controlled by the API response rather than by a trusted local constant, compromise or malicious behavior at the API, its upstream model provider, or an intermediary response-processing component can alter the resource retrieved by the agent. The fixed output path `sunset.png` also overwrites an existing file with that name in the current working directory. This is a limited overwrite primitive because the attacker does not directly control the destination path in the documented command, but it may still destroy an existing file or place unexpected content where an image is expected. ### Attack Path 1. The agent submits an authenticated image-generation request to `api.heybossai.com`. 2. The API or a compromised upstream component returns an attacker-selected value in `image_url`, ...[truncated 1806 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Parse the response strictly and require the selected JSON field to be a nonempty string. 2. Permit only `https` URLs and reject all other schemes. 3. Maintain an explicit allowlist of trusted media-delivery hostnames. 4. Resolve the hostname before connecting and reject loopback, private, link-local, multicast, reserved, and cloud metadata address ranges for both IPv4 and IPv6. 5. Revalidate the scheme, hostname, and resolved address after every redirect, or disable redirects entirely. 6. Configure cURL defensively, for example: - Restrict protocols with `--proto '=https'`. - Restrict redirect protocols with `--proto-redir '=https'`. - Apply `--max-redirs`, `--connect-timeout`, and `--max-time`. - Enforce a maximum accepted response size. - Fail on HTTP errors with `--fail-with-body` or `--fail`. 7. Verify that the response has an expected image MIME type and validate its actual file signature before use. 8. Write to a securely created temporary file, validate it, and then atomically move it to a user-approved destination. 9. Refuse to overwrite an existing output file unless the user explicitly authorizes replacement. 10. Prefer downloading through a trusted application component that implements DNS and redirect checks rather than passing an externally supplied URL directly to a shell command. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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 (38)

Description-Behavior Mismatch

High
Confidence
99% confidence
Finding
The skill is labeled and described as a Brave Search capability, but the content actually exposes a broad third-party orchestration platform with many unrelated capabilities. This kind of scope deception undermines user consent and tool governance because an operator expecting search-only behavior could unknowingly enable email, SMS, scraping, document transfer, and model invocation features.

Context-Inappropriate Capability

High
Confidence
98% confidence
Finding
Email sending and SMS/OTP operations are materially different from web search and content extraction and can trigger real-world external actions. Bundling these capabilities into a search-branded skill increases the chance of unauthorized outreach, spam, social engineering, or account-verification abuse under misleading pretenses.

External Transmission

Medium
Category
Data Exfiltration
Content
One API key, 50+ models across providers (Bedrock, OpenAI, Vertex, ElevenLabs, Replicate, Minimax, and more). Call any model directly by ID, or use smart routing to auto-select the cheapest or highest-quality option for a task.

**Base URL:** `https://api.heybossai.com/v1`
**Auth:** `-H "Authorization: Bearer $SKILLBOSS_API_KEY"`

## List Models
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
One API key, 50+ models across providers (Bedrock, OpenAI, Vertex, ElevenLabs, Replicate, Minimax, and more). Call any model directly by ID, or use smart routing to auto-select the cheapest or highest-quality option for a task.

**Base URL:** `https://api.heybossai.com/v1`
**Auth:** `-H "Authorization: Bearer $SKILLBOSS_API_KEY"`

## List Models
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
One API key, 50+ models across providers (Bedrock, OpenAI, Vertex, ElevenLabs, Replicate, Minimax, and more). Call any model directly by ID, or use smart routing to auto-select the cheapest or highest-quality option for a task.

**Base URL:** `https://api.heybossai.com/v1`
**Auth:** `-H "Authorization: Bearer $SKILLBOSS_API_KEY"`

## List Models
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
One API key, 50+ models across providers (Bedrock, OpenAI, Vertex, ElevenLabs, Replicate, Minimax, and more). Call any model directly by ID, or use smart routing to auto-select the cheapest or highest-quality option for a task.

**Base URL:** `https://api.heybossai.com/v1`
**Auth:** `-H "Authorization: Bearer $SKILLBOSS_API_KEY"`

## List Models
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
One API key, 50+ models across providers (Bedrock, OpenAI, Vertex, ElevenLabs, Replicate, Minimax, and more). Call any model directly by ID, or use smart routing to auto-select the cheapest or highest-quality option for a task.

**Base URL:** `https://api.heybossai.com/v1`
**Auth:** `-H "Authorization: Bearer $SKILLBOSS_API_KEY"`

## List Models
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
One API key, 50+ models across providers (Bedrock, OpenAI, Vertex, ElevenLabs, Replicate, Minimax, and more). Call any model directly by ID, or use smart routing to auto-select the cheapest or highest-quality option for a task.

**Base URL:** `https://api.heybossai.com/v1`
**Auth:** `-H "Authorization: Bearer $SKILLBOSS_API_KEY"`

## List Models
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
## List Models

```bash
curl -s https://api.heybossai.com/v1/models \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY"
```
Confidence
79% confidence
Finding
This documents transmission of requests and an authorization bearer token to an external API endpoint. External transmission is expected for a hosted API integration, but it remains security-relevant because user inputs and credentials leave the local environment and are processed by a third party.

External Transmission

Medium
Category
Data Exfiltration
Content
## List Models

```bash
curl -s https://api.heybossai.com/v1/models \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY"
```
Confidence
79% confidence
Finding
This documents transmission of requests and an authorization bearer token to an external API endpoint. External transmission is expected for a hosted API integration, but it remains security-relevant because user inputs and credentials leave the local environment and are processed by a third party.

External Transmission

Medium
Category
Data Exfiltration
Content
## Chat

```bash
curl -s -X POST https://api.heybossai.com/v1/chat/completions \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
78% confidence
Finding
Chat completions send user prompts to an external aggregator and potentially onward to multiple model providers. In a skill presented as Brave Search, this broad prompt exfiltration path may be unexpected and can expose sensitive content to third parties.

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
The skill exposes image, video, TTS, STT, music, and background-removal functions that are unrelated to the stated Brave Search purpose. While not inherently malicious, this scope expansion weakens least-privilege assumptions and makes it easier to route user data to external providers for unexpected processing.

External Transmission

Medium
Category
Data Exfiltration
Content
## Image Generation

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
84% confidence
Finding
Image-generation requests send prompts and authorization data to an external service, which may further relay content to underlying providers. This is risky if users submit sensitive prompts or if the skill is assumed to be limited to Brave Search rather than broader media processing.

External Transmission

Medium
Category
Data Exfiltration
Content
## Image Generation

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
84% confidence
Finding
Image-generation requests send prompts and authorization data to an external service, which may further relay content to underlying providers. This is risky if users submit sensitive prompts or if the skill is assumed to be limited to Brave Search rather than broader media processing.

External Transmission

Medium
Category
Data Exfiltration
Content
## Video Generation

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
84% confidence
Finding
Video-generation requests transmit prompts and possibly linked media to a third-party API. Because video workflows can include externally hosted source images and substantial data sharing, the absence of warnings and the misleading skill scope make this more dangerous than a clearly labeled media skill.

External Transmission

Medium
Category
Data Exfiltration
Content
## Video Generation

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
84% confidence
Finding
Video-generation requests transmit prompts and possibly linked media to a third-party API. Because video workflows can include externally hosted source images and substantial data sharing, the absence of warnings and the misleading skill scope make this more dangerous than a clearly labeled media skill.

External Transmission

Medium
Category
Data Exfiltration
Content
Image-to-video:

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
80% confidence
Finding
Image-to-video transmits both prompts and externally referenced images to a third-party API, increasing the chance of confidential media being processed off-platform. This is outside the stated purpose of the skill and broadens unexpected data exposure.

External Transmission

Medium
Category
Data Exfiltration
Content
## Text-to-Speech

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
83% confidence
Finding
Text-to-speech requests send user text to an external API and potentially to downstream voice providers. This can expose confidential content and create generated audio without the user fully understanding that third-party processing is involved.

External Transmission

Medium
Category
Data Exfiltration
Content
## Text-to-Speech

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
83% confidence
Finding
Text-to-speech requests send user text to an external API and potentially to downstream voice providers. This can expose confidential content and create generated audio without the user fully understanding that third-party processing is involved.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The skill omits privacy and data-handling warnings for audio, documents, and search content sent to external APIs. Users may provide sensitive files, speech, or queries without realizing this material is transmitted to a third-party aggregation service and possibly downstream model providers.

External Transmission

Medium
Category
Data Exfiltration
Content
## Speech-to-Text

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
90% confidence
Finding
Speech-to-text requests transmit base64-encoded audio and filenames to an external API, which can include highly sensitive personal or business information. In a skill presented as Brave Search, this unexpected audio exfiltration risk is materially concerning.

External Transmission

Medium
Category
Data Exfiltration
Content
## Speech-to-Text

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
90% confidence
Finding
Speech-to-text requests transmit base64-encoded audio and filenames to an external API, which can include highly sensitive personal or business information. In a skill presented as Brave Search, this unexpected audio exfiltration risk is materially concerning.

External Transmission

Medium
Category
Data Exfiltration
Content
## Music Generation

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
74% confidence
Finding
Music-generation requests send prompts externally to third-party providers, which is not aligned with a search skill and may expose proprietary or sensitive creative content. The danger is amplified by the mismatch between the declared purpose and actual capabilities.

External Transmission

Medium
Category
Data Exfiltration
Content
## Background Removal

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
60% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

External Transmission

Medium
Category
Data Exfiltration
Content
## Document Processing

```bash
curl -s -X POST https://api.heybossai.com/v1/run \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
Confidence
89% confidence
Finding
Document processing sends document URLs to an external service for parsing or extraction, which can expose sensitive files and their contents to third parties. For a skill marketed as Brave Search, this is an unexpected and potentially high-sensitivity data transfer path.

Static analysis

No suspicious patterns detected.