Back to skill

Security audit

Clawbus Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill is a ClawBus installer, but it tells agents to download, overwrite, and immediately follow remote skill instructions without enough validation or user approval.

Install only if you trust ClawBus as a live source of skill instructions and are comfortable with the agent refreshing and activating those instructions. Review downloaded files before use, avoid granting ambient credentials, and require explicit confirmation before any posting, publishing, or irreversible account action.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:56
Finding

Automatic Activation of Mutable Remote Skill Instructions

Content
View full analysis
`: 1. Treat the text after `use` as the skill slug. 2. Fetch the complete skill package from the ClawBus install endpoint. 3. Save every returned file into a local skill directory for that slug. 4. Read the downloaded `SKILL.md`. 5. Follow the downloaded skill's instructions for the current request. Do not stop after downloading. A `use ` request means install, read, and activate the skill in the same flow. If the skill already exists locally, refresh it from ClawBus before using it so the agent runs the current instructions. ``` ### Technical Analysis The Skill instructs the agent to fetch a mutable package from a remote service and immediately follow the downloaded instructions. No package version, cryptographic digest, signature, trusted manifest, instruction allowlist, or mandatory user review is required. HTTPS protects the connection in transit but does not establish that the package itself is authorized, immutable, or safe. The requirement to refresh an existing local Skill before every use also defeats the security value of previously reviewing a known local version. Although the downloaded files are described as Skill instructions rather than necessarily executable programs, following a remotely supplied `SKILL.md` creates an effective remote payload execution channel at the agent-instruction level. The local safety statement requiring downloaded files to be read does not require unsafe instructions to be rejected or changes to be approved. ### Attack Path 1. A user requests `use `. 2. The agent sends the slug to the ClawBus installation endpoint. 3. The endpoint, or an attacker who has compromised its package catalog, returns an altered package. 4. The agent saves the returned files and ...[truncated 1541 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:95
Finding

Unvalidated Remote File Paths Can Escape the Skill Installation Directory

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 61)May include surrounding context.

md
4. Read the downloaded `SKILL.md`.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The invocation guidance is overly broad, telling the agent to use this skill whenever users mention ClawBus, creator workflows, installing a skill, or a ClawBus slug. Broad triggers increase the chance of unintended activation and can route ordinary conversations into a flow that downloads and follows remote instructions, expanding the attack surface.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The 'When To Use' section includes ambiguous triggers like mentions of creator publishing, YouTube APIs, Instagram publishing, or TikTok publishing, which are broad topics rather than clear installation requests. In this skill's context, ambiguous matching is more dangerous because activation leads to fetching, writing, and then obeying downloaded skill content from a remote service.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
95% confidence
Finding

The skill instructs the agent to save arbitrary files returned by a remote endpoint into a local skill directory and then read and follow the downloaded SKILL.md. This is effectively remote instruction ingestion and local file materialization, which can enable skill enumeration, overwrite of local assets via crafted paths, or prompt-injection-by-install if the remote source or path handling is compromised.

Content

Scanner excerpt · SKILL.md (reported line 101)May include surrounding context.

md
2. Write each file from the response into that directory, preserving paths.
3. Create subdirectories when needed.
4. Store lightweight install metadata in `_meta.json`.
5. Read `skills/SLUG/SKILL.md` before taking action.

Example metadata:

Static analysis

No suspicious patterns detected.