Back to skill

Security audit

Howtoletmyagent Installer

Security checks for vulnerabilities and agentic risk

Overview

This skill has a legitimate installer purpose, but it asks the agent to trust remote manifests for commands and persistent skill files with too little validation.

Only install this if you fully trust the manifest source and can review the exact command and file list before approval. Prefer a vetted ClawHub install path, avoid approving arbitrary preview domains, and do not use it for manifests that include unexpected commands, scripts, broad file paths, or unclear package identity.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (3)

T03 · Remote Payload Retrieval and Execution

Error
Location
SKILL.md:18
Finding

Manifest-Controlled Remote Command Execution

Content
View full analysis
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:23
Finding

Persistent Installation of Unvalidated Remote Skill Content

Content
View full analysis
/` using the manifest `files` array. Success condition: - The requested companion skill is installed. - The user gets the exact prompt they can use next to trigger it. ``` ### Technical Analysis The fallback workflow writes remotely supplied files directly into the persistent `skills//` directory and then instructs the user to reload OpenClaw. No requirement is defined to validate the contents, constrain file types, canonicalize destination paths, reject traversal sequences or symbolic links, verify file hashes, inspect executable content, or authenticate the manifest. Because skill files can define instructions that affect agent behavior, activating unreviewed remote files can introduce attacker-controlled goals, unsafe tool-use instructions, or attempts to weaken safety constraints. If the manifest can include scripts or other executable resources, the same installation mechanism may also place executable payloads into the skill package. ### Attack Path 1. An attacker controls or modifies a manifest returned by an accepted source. 2. The manifest's `files` array contains malicious skill instructions, executable content, or unsafe file paths. 3. A ClawHub installation is unavailable or made to fail, activating the local-file fallback. 4. The agent writes each remote file into the workspace skill directory exactly as provided. 5. The user starts a new session or restarts the gateway. ...[truncated 645 chars]
Remediation
View remediation
/` directory. - Reject absolute paths, `..` traversal, alternate path separators, symbolic links, hard links, device files, and other special filesystem objects. - Restrict installation to an explicit allowlist of documented, non-executable skill file types. - Reject scripts, binaries, archives, and executable permissions unless they pass a separate trusted-code review process. - Statically inspect skill instructions for attempts to override system constraints, capture secrets, invoke unrelated tools, or install additional payloads. - Present the complete normalized file list and exact content diff to the user before writing anything. - Stage files in an isolated directory, validate the complete package, and install atomically only after validation succeeds. - Define rollback and removal procedures so an unsafe skill can be disabled without loading it. ]]>

T08 · Insecure Dependencies

Error
Location
SKILL.md:16
Finding

Undefined Trust Expansion to User-Approved Manifest Hosts

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (2)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill allows invocation from a vague natural-language request such as 'learn the skill from this page' when a URL is merely present in context. That broad trigger can cause the agent to fetch and potentially install untrusted or unintended content based on ambiguous page context, increasing the risk of prompt-driven or context-confusion attacks.

Content

No source excerpt is available for this finding.

Unrestricted Tool Access

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill grants unrestricted tool access without appropriate constraints. An agent with unfettered tool access can perform arbitrary actions including file modification, network requests, and code execution.

Content

Scanner excerpt · SKILL.md (reported line 31)May include surrounding context.

md
Safety rules:

- Only trust manifests from https://howtoletmyagent.xyz unless the user explicitly approves another domain.
- Never execute arbitrary code from the manifest. This manifest is for skill files, not shell scripts.
- Show the target install path before writing files.
- If a manifest looks malformed or unsafe, stop and explain the issue instead of guessing.