Back to skill

Security audit

ssh-agentd-control

Security checks for vulnerabilities and agentic risk

Overview

The skill is coherent for administering ssh-agentd, but it gives an agent privileged service-control, boot-persistence, and remote-command examples without enough scoping or confirmation guidance.

Install only if you intentionally manage this exact ssh-agentd setup. Before using the mutating commands, inspect ssh-agentd.service and the referenced binary/config ownership and permissions, keep the API bound to localhost unless needed, protect SSH_AGENTD_TOKEN, and require explicit approval before enabling, restarting, stopping, or invoking /run.

Vulnerability Patterns
  • System PersistenceInstalls backdoors, hooks, services, or scheduled tasks that survive the run
  • 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
Findings (1)

T06 · System Persistence

Error
Location
SKILL.md:27
Finding

Privileged Systemd Service Persistence Without Verifiable Service Definition

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 27
Vulnerability Type: T06: System Persistence
Risk Level: High

Complete Code Snippet:

bash
sudo systemctl daemon-reload
sudo systemctl enable --now ssh-agentd.service
sudo systemctl restart ssh-agentd.service
sudo systemctl stop ssh-agentd.service

Technical Analysis

The command sudo systemctl enable --now ssh-agentd.service performs two privileged operations:

  1. It starts the service immediately.
  2. It registers the service to start automatically on future boots.

Persistence is related to the Skill's declared support for a systemd-resident SSH agent and persistent SSH sessions. However, enabling boot persistence is not necessary for routine API requests, health checks, session inspection, or one-time service execution. It therefore exceeds the minimum privileges required for those common operations unless the user explicitly requests automatic startup.

The service unit, referenced executable, and configuration are not included in the audited project. The documentation instead references external resources such as /home/krex/.openclaw/workspace-hermes/ssh-agentd/bin/ssh-agentd and /home/krex/.openclaw/ssh-agentd/hosts.yaml. Consequently, this package cannot establish what executable systemd will launch, which user it will run as, or whether appropriate systemd sandboxing is configured.

Attack Path

  1. An attacker or another compromised process creates or modifies ssh-agentd.service, its referenced executable, or associated configuration outside this project.
  2. An operator or AI agent follows the Skill instructions and authorizes the sudo systemctl enable --now ssh-agentd.service command.
  3. systemd immediately launches the externally defined and unaudited service with the privileges configured by its unit.
  4. The enable operation installs cross-session persistence, causing the same service to launch after subseque ...[truncated 857 chars]
Remediation
View remediation

Remediation Suggestions

  • Do not enable the service automatically as part of routine Skill use. Separate API operations, temporary service startup, and boot-persistence setup into distinct workflows.
  • Require explicit, informed user confirmation before executing sudo systemctl enable, clearly stating that it creates persistence across reboots.
  • Prefer systemctl start ssh-agentd.service without enable when only temporary operation is required.
  • Before startup or enablement, inspect the effective unit with systemctl cat ssh-agentd.service and systemctl show ssh-agentd.service.
  • Verify the referenced executable and configuration using trusted ownership, restrictive permissions, and a known-good checksum or signature.
  • Confirm that neither the unit nor its executable is writable by untrusted users.
  • Run the daemon as a dedicated unprivileged account, or use a per-user systemd service where system-wide privileges are unnecessary.
  • Harden the unit with appropriate controls such as NoNewPrivileges=true, ProtectSystem=strict, ProtectHome=true, PrivateTmp=true, restricted capabilities, and narrowly scoped filesystem access.
  • Keep the API bound to loopback unless remote exposure is explicitly required, and retain authentication for commands capable of remote execution.
  • Include the service unit or a verifiable unit template in the audited package so its executable path, service user, capabilities, and sandboxing can be reviewed.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (13)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
87% confidence
Finding

The skill invokes shell commands with privileged service management and API operations, but it declares no explicit tool scope or permissions boundary. That increases the chance the agent may overreach and execute impactful commands in contexts where the user only asked for general troubleshooting or information.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The activation text is broad enough to trigger on common troubleshooting phrases like metrics, session status, or startup behavior, which could invoke a skill capable of sudo service control and remote command execution. Overbroad routing is dangerous because it may cause high-impact operational actions to be suggested or performed in situations that only needed low-risk diagnostic help.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 17)May include surrounding context.

快速检查

bash
sudo systemctl is-enabled ssh-agentd.service
sudo systemctl is-active ssh-agentd.service
ss -ltnp | grep 18081
sudo systemctl status ssh-agentd.service --no-pager -l | sed -n '1,80p'

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 18)May include surrounding context.

快速检查

bash
sudo systemctl is-enabled ssh-agentd.service
sudo systemctl is-active ssh-agentd.service
ss -ltnp | grep 18081
sudo systemctl status ssh-agentd.service --no-pager -l | sed -n '1,80p'

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 20)May include surrounding context.

快速检查

bash
sudo systemctl is-enabled ssh-agentd.service
sudo systemctl is-active ssh-agentd.service
ss -ltnp | grep 18081
sudo systemctl status ssh-agentd.service --no-pager -l | sed -n '1,80p'

Session Persistence

Medium
Category
Rogue Agent
Confidence
80% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 17)May include surrounding context.

快速检查

bash
sudo systemctl is-enabled ssh-agentd.service
sudo systemctl is-active ssh-agentd.service
ss -ltnp | grep 18081
sudo systemctl status ssh-agentd.service --no-pager -l | sed -n '1,80p'

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The examples show service enable/restart/stop and a remote command execution endpoint without prominent warnings that these actions can alter host state, affect availability, or execute commands on remote systems. In a skill meant for agent use, cookbook-style examples can be operationalized directly, making omission of safety gates materially risky.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
83% confidence
Finding

sudo systemctl daemon-reload performs a privileged operational change and can apply altered unit definitions. In a skill lacking strict scoping and confirmation rules, including this as a ready-to-run example raises the risk of unintended privileged changes during routine troubleshooting.

Content

Scanner excerpt · SKILL.md (reported line 26)May include surrounding context.

启停与自启

bash
sudo systemctl daemon-reload
sudo systemctl enable --now ssh-agentd.service
sudo systemctl restart ssh-agentd.service
sudo systemctl stop ssh-agentd.service

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
92% confidence
Finding

sudo systemctl enable --now ssh-agentd.service both persists the service across reboots and starts it immediately, creating a durable privileged change. If triggered inappropriately, it can establish unwanted persistence and expose the local API service without sufficient review of configuration or access controls.

Content

Scanner excerpt · SKILL.md (reported line 27)May include surrounding context.

bash
sudo systemctl daemon-reload
sudo systemctl enable --now ssh-agentd.service
sudo systemctl restart ssh-agentd.service
sudo systemctl stop ssh-agentd.service

Session Persistence

Medium
Category
Rogue Agent
Confidence
94% confidence
Finding

This command explicitly enables the service at boot, which is a persistence mechanism. In a skill that also describes API access and remote execution, persistence materially increases risk because it makes the service continuously available and harder to notice if enabled without deliberate intent.

Content

Scanner excerpt · SKILL.md (reported line 27)May include surrounding context.

bash
sudo systemctl daemon-reload
sudo systemctl enable --now ssh-agentd.service
sudo systemctl restart ssh-agentd.service
sudo systemctl stop ssh-agentd.service

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
86% confidence
Finding

sudo systemctl restart ssh-agentd.service is a privileged, state-changing action that can disrupt active sessions or alter service behavior after config changes. In an auto-invoked skill, restart examples without safety gates increase the chance of accidental service interruption.

Content

Scanner excerpt · SKILL.md (reported line 28)May include surrounding context.

bash
sudo systemctl daemon-reload
sudo systemctl enable --now ssh-agentd.service
sudo systemctl restart ssh-agentd.service
sudo systemctl stop ssh-agentd.service

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
84% confidence
Finding

sudo systemctl stop ssh-agentd.service can immediately terminate service availability and any dependent workflows. Including it as an unguarded example is risky because a troubleshooting flow could unnecessarily turn into a denial-of-service against the managed agent.

Content

Scanner excerpt · SKILL.md (reported line 29)May include surrounding context.

sudo systemctl daemon-reload sudo systemctl enable --now ssh-agentd.service sudo systemctl restart ssh-agentd.service sudo systemctl stop ssh-agentd.service

text

## API 调用要点

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The script reads a credential-like value from SSH_AGENTD_TOKEN and includes it as an Authorization header in outgoing HTTP requests. There is no confirmation prompt, logging, comment, or docstring warning the user that authentication data may be sent to the target endpoint.

Content

No source excerpt is available for this finding.