Back to skill

Security audit

Agent Avatar Manager

Security checks for vulnerabilities and agentic risk

Overview

The skill has a legitimate avatar-management purpose, but it uses unsafe shell templates, unrestricted URL downloads, and plaintext API-key persistence that users should review before installing.

Review this skill before installing. Use a direct local image when possible, avoid pasting sensitive descriptions or private URLs, do not store a valuable API key in TOOLS.md unless you accept plaintext storage, and prefer pinned install sources. The skill should validate agent names, agent IDs, filenames, and URLs before any curl or openclaw command is run.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (4)

T08 · Insecure Dependencies

Error
Location
README.md:21
Finding

Unpinned packages and mutable sources are executed during installation

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:112
Finding

Shell command injection through unvalidated identity-derived values

Content
View full analysis
/dev/null | grep -oE "^{AgentName}([0-9]+)\.(jpg|png)$" | grep -oE "[0-9]+" | sort -n | tail -1 ``` ```bash openclaw agents set-identity --agent {AGENT_ID} --avatar "avatars/{FILENAME}" ``` The affected values are derived according to the following instructions: ```text Read the current agent's `IDENTITY.md` and extract: - **Agent name** (used for filenames, e.g. `Ashley`) - **Agent id** (used by `openclaw agents set-identity`, e.g. `main`, `zhima`) ``` ### Technical Analysis `AgentName` and `AGENT_ID` are obtained from `IDENTITY.md` or runtime context and then interpolated into shell commands without strict validation. Placing `AgentName` inside double quotes does not make arbitrary substitution safe. Shell constructs such as command substitution can still be evaluated if an implementation builds the command as a string after replacing the placeholder. `AGENT_ID` is not quoted at all, so shell metacharacters, whitespace, substitutions, and additional arguments may alter the command. `FILENAME` is also derived from the agent name and is subsequently used in shell commands. Therefore, validating only the numeric suffix would not eliminate the injection risk. ### Attack Path 1. An attacker gains the ability to modify `IDENTITY.md`, supplies a crafted identity file, or otherwise influences the agent name or ID. 2. The malicious value contains shell syntax, such as command substitution or a command separator. 3. The user invokes the avatar-management skill. 4. The agent follows the documented template and constructs a shell command containing the untrusted value. 5. The shell interprets the injected syntax and executes the attacker's command under the agent process's user account. ### Impact Assessment Exploitation could provide arb ...[truncated 380 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:105
Finding

Unrestricted image URL retrieval permits SSRF and unsafe local resource access

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:32
Finding

Freepik API key is persistently stored in a plaintext workspace file

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (9)

Rp1

Medium
Category
MCP Rug Pull
Confidence
91% confidence
Finding

The README instructs users to execute an unpinned package via npx, which fetches and runs whatever version is current at install time. If the package is compromised, updated maliciously, or a dependency is hijacked, users could execute attacker-controlled code during installation.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
91% confidence
Finding

The clawhub install command is also unpinned, so users are directed to run a remote package manager command without version locking. This creates a supply-chain risk because future package changes or account compromise could alter what gets executed on user systems.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README says a pasted Freepik API key will be auto-saved to TOOLS.md for future use, but it does not clearly warn users that this persists a secret to a workspace file. Storing credentials in plaintext documentation-style files can lead to accidental disclosure through source control, file sharing, backups, or other tools that read workspace content.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill description says it auto-saves to the workspace and updates IDENTITY.md, but it does not present this as a user-facing warning or require explicit confirmation before making persistent configuration changes. This can surprise users and lead to unintended modification of agent identity state.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger phrases are very broad and include common requests like 'change avatar', which can cause the skill to activate in contexts where the user did not clearly consent to file writes, identity changes, or external API use. Because the skill performs persistent modifications and may contact third-party services, overly loose activation increases the chance of unintended execution.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The description does not clearly warn users that style descriptions are sent to Freepik during search mode and that remote image URLs may be fetched from third parties. This creates a privacy and data-handling risk because user input and network metadata may be disclosed externally without sufficiently explicit notice.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The skill sends constructed search queries to Freepik and uses an API key read from local configuration, creating clear external data transmission. If user descriptions contain sensitive details, those details are disclosed to a third party; additionally, the skill encourages automatic network access without an explicit consent gate tied to the transmission step.

Content

Scanner excerpt · SKILL.md (reported line 76)May include surrounding context.

Step 3 — Call Freepik API

bash
curl -s "https://api.freepik.com/v1/resources?locale=en&page=1&limit=10&order=relevance&term={QUERY}&filters%5Bcontent_type%5D%5Bvector%5D=1&filters%5Blicense%5D%5Bfree%5D=1" \
  -H "x-freepik-api-key: {API_KEY}" \
  -H "Accept: application/json"

Session Persistence

Medium
Category
Rogue Agent
Confidence
91% confidence
Finding

The skill persists downloaded content into the user's workspace and then updates agent identity to reference it, making the change durable across sessions. Persistent writes are not inherently unsafe, but here they increase risk because remote content is saved locally and becomes part of ongoing agent configuration, potentially without strong validation or clear consent.

Content

Scanner excerpt · SKILL.md (reported line 128)May include surrounding context.

Step 6 — Download and save

bash
mkdir -p ~/.openclaw/workspace/avatars
curl -sL -o ~/.openclaw/workspace/avatars/{FILENAME} "{CHOSEN_URL}"

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The instructions state that IDENTITY.md info and the user's description should be translated to English for search construction. This imposes a language choice automatically rather than offering the user a choice or explaining the constraint as a user-visible, justified locale requirement.

Content

No source excerpt is available for this finding.