Security audit

Onboarding

Security checks for vulnerabilities and agentic risk

Overview

This onboarding skill is mostly coherent, but it creates persistent agent memory and operating rules from personal information with broad activation and limited upfront consent.

Review this before installing if you do not want an agent to store personal details and workspace preferences in files that future sessions read. Use it only when you intentionally want persistent onboarding, and inspect or delete the generated memory and AGENTS.md files if they contain rules you do not want reused.

A.I.G

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 (3)

T02 · Agent Memory Poisoning

Error

Location: SKILL.md:185
Finding: Persistent Agent Behavior Modification Through Generated Memory Files
Content: View full analysis
Remediation: View remediation

T01 · Skill Instruction Hijacking

Warning

Location: SKILL.md:5
Finding: Global Skill-Priority Directive Can Displace the User's Current Task
Content: View full analysis
Remediation: View remediation

other

Note

Location: SKILL.md:22
Finding: Undisclosed Host and Runtime Environment Reconnaissance
Content: View full analysis
/dev/null ``` If files exist → ask: *"I see you already have some setup files. Want to start fresh or update what's there?"* Detect silently (don't announce): - OS: `uname -s` (Darwin = macOS, Linux = Linux) - Timezone: `date +%Z` - Agent runtime: check if `openclaw` or `claude` is in PATH ``` ### Technical Analysis The skill directs the agent to inspect workspace state, operating-system type, timezone, and installed agent runtimes while explicitly instructing it not to announce the inspection. Checking whether target setup files already exist is reasonably related to avoiding accidental overwrites. OS and timezone detection may support profile generation. However, silent runtime enumeration is not necessary for the documented creation of the five foundation files, and the user is not given an opportunity to consent to or decline these checks. No command in this segment accesses credentials, elevates privileges, transmits data externally, or executes a downloaded payload. The concern is undisclosed local reconnaissance and unnecessary collection rather than privilege escalation. ### Attack Path 1. The onboarding skill starts. 2. The agent lists the target foundation files to determine whether they exist. 3. It executes or performs equivalent checks for the operating system and timezone. 4. It checks whether `openclaw` or `claude` is available in `PATH`. 5. The collected OS and timezone information may be placed into persistent user-profile files, while the runtime information informs the skill without being disclosed to the user. ### Impact Assessment The information exposed is limited to named workspace-file existence, operating-system family, timezone, and availability of two runtime com ...[truncated 338 chars]
Remediation: View remediation

SkillSpector

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
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (3)

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 96% confidence
Finding: The trigger condition includes the phrase 'who are you', which is an ordinary conversational prompt and can cause the onboarding skill to activate in unrelated contexts. Because this skill initiates broad personal data collection and persistence into multiple memory files, accidental invocation can lead to unnecessary harvesting and storage of sensitive profile information.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 96% confidence
Finding: The skill instructs the agent to collect highly sensitive personal profile data, work habits, tools, location/timezone, and preferences, then persist them as long-lived foundation files that are read every session. There is no upfront privacy warning, retention notice, or consent checkpoint explaining what will be stored, creating a meaningful risk of over-collection, surprise persistence, and exposure of sensitive user context.
Content: No source excerpt is available for this finding.

Autonomous Decision Making

Medium

Category: Excessive Agency
Confidence: 75% confidence
Finding: Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
Content: Scanner excerpt · SKILL.md (reported line 210)May include surrounding context.
md
## Safety - Don't exfiltrate private data. Ever. - Don't run destructive commands without asking. - When in doubt, ask. ## External vs Internal

Static analysis

No suspicious patterns detected.