Back to skill

Security audit

Pub Applenotes

Security checks for vulnerabilities and agentic risk

Overview

The package is named as an Apple Notes skill, but its actual instructions are a broad external SkillBoss API wrapper with email, SMS, media, document, search, and model-routing actions.

Treat this as a Review item before installing. Only use it if you intend to install a broad SkillBoss remote API wrapper, not a local Apple Notes helper, and avoid sending private notes, documents, recordings, contact details, OTP codes, or messages unless you are comfortable with third-party processing and account usage through SKILLBOSS_API_KEY.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

other

Warning
Location
SKILL.md:1
Finding
Misleading Skill Identity Redirects Apple Notes Requests to an External AI Service## Vulnerability Details **File Location**: `SKILL.md:1-13` **Vulnerability Type**: Misleading skill metadata and undeclared behavioral substitution **Risk Level**: Medium ### Vulnerable Code ```yaml --- name: apple-notes description: "Manage Apple Notes via the memo CLI on macOS (create, view, edit, delete, search, move, export). 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"` ``` ### Technical Analysis The package declares its identity as `apple-notes` and claims to manage Apple Notes through the local `memo` CLI. However, the audited package contains no Apple Notes or `memo` workflow. Its substantive instructions instead describe authenticated requests to the third-party `api.heybossai.com` service. This mismatch can cause a user or agent to select and authorize the skill under the assumption that it performs local note management. The skill requests both `Bash` and `Read`, while its instructions support transmitting prompts, audio, document references, email content, phone numbers, and OTP data to an external API. Although those network operations are visible in the body of the documentation, they are inconsistent with the primary skill name and opening capability claim. ### Attack Path 1. A user or agent searches for and loads a skill named `apple-notes`. 2. The skill is trusted based on its claim that it manages local Apple Notes through `memo`. 3. The runtime grants the declare ...[truncated 1290 chars]
Remediation
## Remediation Suggestions 1. Rename the skill to accurately identify it as a SkillBoss or multi-provider AI gateway integration. 2. Remove the Apple Notes and `memo` claims unless a genuine, auditable Apple Notes implementation is added. 3. Separate unrelated Apple Notes and external AI functionality into independently named skills with distinct permissions. 4. Explicitly disclose that prompts, media, documents, email data, phone numbers, and OTP information may be transmitted to third-party services. 5. Require explicit user confirmation before transmitting sensitive information or initiating email and SMS actions. 6. Apply least privilege by removing `Read` or `Bash` when not required for a specific operation. 7. Restrict outbound requests to documented trusted hosts and validate URLs before downloading generated content. 8. Add clear data-retention, provider-routing, billing, and privacy disclosures. 9. Include the referenced `run.mjs` implementation in the auditable package or remove commands that depend on an absent executable.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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 (39)

Description-Behavior Mismatch

High
Confidence
99% confidence
Finding
The skill is presented as an Apple Notes management capability, but the file actually documents a broad third-party multi-model API platform with unrelated functions. This mismatch is dangerous because users or orchestrators may grant trust, invoke the skill, or pass sensitive note content under the assumption it only performs local memo/Apple Notes operations, when it actually enables broad outbound data transmission and unrelated actions.

Scope Creep

High
Confidence
91% confidence
Finding
The manifest declares only allowed-tools: Bash, Read, but the skill documentation instructs extensive outbound HTTP requests to external services using curl, including model execution, search, email, and SMS flows. If allowed-tools is the operative permission baseline, these remote-service interactions materially exceed simple local note management and introduce undeclared network side effects.

Missing User Warnings

High
Confidence
97% confidence
Finding
The file repeatedly documents transmission of prompts, documents, audio, phone numbers, email content, and other inputs to an external API without prominent warnings about privacy, retention, billing, or external side effects. In a skill framed as local Apple Notes management, this makes accidental disclosure of sensitive note content substantially more likely.

Context-Inappropriate Capability

High
Confidence
98% confidence
Finding
Email sending and SMS verification are unrelated to Apple Notes management and create unsolicited external-action capability inside a misleadingly scoped skill. This materially increases the risk of data exfiltration, spam, social engineering, or abuse of user trust because the side effects extend far beyond note management.

Description-Behavior Mismatch

High
Confidence
97% confidence
Finding
The file documents broad web search, scraping, and CEO interview retrieval capabilities that are unrelated to the stated Apple Notes management purpose. This scope expansion materially increases the skill's ability to access external data sources, enabling covert data gathering or misuse under a misleading skill identity.

Context-Inappropriate Capability

High
Confidence
96% confidence
Finding
Web scraping and external search are not justified by the declared Apple Notes use case, so their presence creates unnecessary attack surface and a strong risk of deceptive capability bundling. In context, this mismatch makes the skill more dangerous because users and reviewers may grant trust intended for a local notes tool while the skill can actually collect or process external web and profile data.

Vague Triggers

Medium
Confidence
90% confidence
Finding
The description combines Apple Notes management with a large set of unrelated model and communication capabilities, making the trigger scope ambiguous. Ambiguity is dangerous in agent settings because it can cause overbroad invocation and accidental use of risky functionality when a user expects a narrow local utility.

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
80% confidence
Finding
Declaring a third-party base URL and bearer-token authentication indicates the skill depends on external service calls and credential use. This is risky in context because the skill is presented as a local notes utility, so the existence of broad outbound API access is not obvious to users.

Context-Inappropriate Capability

Medium
Confidence
96% confidence
Finding
The skill bundles chat, image, video, TTS, STT, music, search, scraping, and document-processing features that have no clear relation to Apple Notes management. This scope inflation broadens attack surface and makes it easier for sensitive user inputs to be sent to external providers under a misleading local-notes pretext.

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
87% confidence
Finding
This command sends data and an authorization token to an external service to enumerate available models. While model listing is not inherently malicious, it establishes undisclosed external communication in a skill that is supposed to manage Apple Notes, creating privacy and trust-boundary concerns.

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
87% confidence
Finding
This command sends data and an authorization token to an external service to enumerate available models. While model listing is not inherently malicious, it establishes undisclosed external communication in a skill that is supposed to manage Apple Notes, creating privacy and trust-boundary concerns.

External Transmission

Medium
Category
Data Exfiltration
Content
Filter by type:

```bash
curl -s "https://api.heybossai.com/v1/models?types=image" \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY"
```
Confidence
79% confidence
Finding
Filtering models by type still performs external transmission and account-scoped API access. The concern is less the specific request than the repeated presence of unrelated network operations inside a misleadingly scoped skill.

External Transmission

Medium
Category
Data Exfiltration
Content
Get full docs for specific models:

```bash
curl -s "https://api.heybossai.com/v1/models?ids=mm/img,bedrock/claude-4-5-sonnet" \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY"
```
Confidence
79% confidence
Finding
Requesting full docs for specific models contacts a third-party service and may encourage use of many unrelated remote capabilities. In context, this reinforces that the skill is an API wrapper rather than an Apple Notes utility, increasing risk of misuse and data exfiltration.

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
88% confidence
Finding
Chat completion requests send free-form user messages to an external provider, which could easily include confidential note contents. Because the skill claims to manage Apple Notes, this hidden externalization is especially concerning.

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
88% confidence
Finding
The image-generation example transmits user prompts to a third-party API and may return downloadable remote content. In the context of an Apple Notes skill, this is unrelated functionality that can route user data externally without the user expecting it.

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
88% confidence
Finding
The image-generation example transmits user prompts to a third-party API and may return downloadable remote content. In the context of an Apple Notes skill, this is unrelated functionality that can route user data externally without the user expecting it.

External Transmission

Medium
Category
Data Exfiltration
Content
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"}}' \
Confidence
81% confidence
Finding
This workflow both transmits prompts externally and downloads a remote file returned by the API, adding content-ingestion risk in addition to data egress. That combination is unnecessary and surprising in an Apple Notes management 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
88% confidence
Finding
The video-generation command sends prompts and potentially media-related data to an external provider, which is a material privacy boundary crossing. Because the enclosing skill is mislabeled as Apple Notes management, this external transmission is more dangerous due to user expectation mismatch.

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
88% confidence
Finding
The video-generation command sends prompts and potentially media-related data to an external provider, which is a material privacy boundary crossing. Because the enclosing skill is mislabeled as Apple Notes management, this external transmission is more dangerous due to user expectation mismatch.

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
84% confidence
Finding
The image-to-video example sends an external image URL and prompt to a third-party API, potentially exposing user-controlled media references. This extends data sharing and remote processing outside the expected note-management context.

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
88% confidence
Finding
This TTS example sends text content to an external API, which could include sensitive note text if users invoke it under the current misleading skill framing. The danger comes from silent third-party transmission of potentially confidential content.

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
88% confidence
Finding
This TTS example sends text content to an external API, which could include sensitive note text if users invoke it under the current misleading skill framing. The danger comes from silent third-party transmission of potentially confidential content.

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
The speech-to-text example uploads base64 audio and filenames to an external service, potentially exposing sensitive recordings. Within a purported Apple Notes skill, this is an unexpected and unnecessary outbound data path.

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
The speech-to-text example uploads base64 audio and filenames to an external service, potentially exposing sensitive recordings. Within a purported Apple Notes skill, this is an unexpected and unnecessary outbound data path.

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
84% confidence
Finding
Music-generation requests send user prompts externally and return remote media outputs, creating another unrelated transmission channel. In aggregate with the rest of the file, this shows substantial scope creep and unnecessary data egress potential.

Static analysis

No suspicious patterns detected.