Back to skill

Security audit

TM Robot Control

Security checks for vulnerabilities and agentic risk

Overview

This is a real robot-control skill with a clear purpose, but it includes live-motion scripts and tests that can move connected hardware without adequate safety gates.

Review before installing on any machine that can reach a real robot. Treat the examples and tests as live hardware programs, remove or gate module-scope motion scripts, require explicit operator confirmation and independent safety interlocks, validate motion/output bounds, fix reset_alarm semantics, and pin reviewed dependency versions before use.

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)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/move_x_minus150.py:5
Finding

Robot Motion Executes Automatically When Modules Are Imported or Discovered

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
tm_robot.py:349
Finding

Motion and I/O APIs Forward Unvalidated Safety-Critical Parameters

Content
View full analysis
bool: if not self._sct_conn: logger.error("SCT not connected") print("[错误] 未建立 SCT 连接") return False try: await self._sct_conn.move_to_joint_angles_ptp( joint_angles_goal=joints, speed_perc=speed, acceleration_duration=accel ) logger.info(f"Move joint: {joints}") print(f"[OK] 关节运动:{[round(j, 1) for j in joints]} deg") return True ``` `tm_robot.py:390-400`: ```python if not self._sct_conn: logger.error("SCT not connected") print("[错误] 未建立 SCT 连接") return False try: await self._sct_conn.move_to_relative_point_line( relative_point_goal=delta, speed=speed, acceleration_duration=accel ) logger.info(f"Move relative line: {delta}") print(f"[OK] 直线运动:dX={delta[0]:.1f}, dY={delta[1]:.1f}, dZ={delta[2]:.1f} mm") return True ``` `tm_robot.py:457-469`: ```python if not self._sct_conn: logger.error("SCT not connected") print("[错误] 未建立 SCT 连接") return False try: await self._sct_conn.move_on_circle( tcp_point_1=point1, tcp_point_2=point2, speed=speed, acceleration_duration=accel, arc_angle=arc_angle ) logger.info(f"Move circle: via={point1}, end={point2}") print(f"[OK] 圆弧运动:中间点 {point1[:3]}, 终点 {point2[:3]}") return True ``` `tm_robot.py:492-510`: ```python async def set_digital_output(self, pin: int, value: bool) -> bool: if not self._sct_conn: logger.error("SCT not connected") return False try: if hasattr(self._sct_conn, 'set_output'): ...[truncated 2878 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
tm_robot.py:476
Finding

Alarm Reset API Does Not Reset Alarms and Can Mislead Safety-Critical Callers

Content
View full analysis
bool: """停止当前运动""" if not self._sct_conn: return False try: await self._sct_conn.stop_motion() logger.info("Motion stopped") return True except Exception as e: logger.error(f"Stop motion failed: {e}") return False async def reset_alarm(self) -> bool: """复位报警状态""" return await self.stop_motion() ``` ### Technical Analysis The method named `reset_alarm()` does not issue an alarm acknowledgement or reset operation. It delegates directly to `stop_motion()` and returns that operation's success value. A successful return therefore means only that the stop command completed. It does not establish that an alarm was cleared, that the robot is ready, or that subsequent motion is safe. This conflicts with the declared safety capability in project metadata and documentation and can cause automation code or an operator interface to interpret `True` as evidence of a successful alarm reset. Safety-related APIs must provide precise postconditions. Aliasing alarm reset to motion stop creates an unsafe state-confusion condition. ### Attack Path 1. The robot enters an alarm or fault state. 2. An application calls `reset_alarm()` as part of recovery logic. 3. The method calls `stop_motion()` and returns `True` if stopping succeeds. 4. The application interprets `True` as confirmation that the alarm was reset. 5. Recovery logic proceeds using an incorrect model of the robot state, potentially attempting movement or re-enabling related equipment without resolving the underlying fault. No attacker is necessary; normal recovery automation can trigger the hazardous behavior. An attacker able to induce or time a fault could increase the likelihood of unsafe state transitions. ### ...[truncated 472 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
manifest.json:12
Finding

Third-Party Robot Control Dependency Is Unpinned and Unverified

Content
View full analysis
=1.0.0", "asyncio": "*" }, ``` `SKILL.md:17-20`: ```markdown ## Requirements - Python >= 3.6 - techmanpy >= 1.0.0 ``` ### Technical Analysis The robot-control package accepts any future `techmanpy` release at or above version `1.0.0`. No upper bound, exact lock, package hash, or reviewed artifact source is supplied. The manifest also declares `asyncio` with a wildcard even though `asyncio` is part of the Python standard library in the supported runtimes. Installing a separately published package with that name unnecessarily expands the dependency and name-confusion attack surface. Because `techmanpy` is imported inside safety-critical connection and actuation methods, the selected package executes with the same local privileges as the skill and receives robot addresses and motion commands. A compromised or incompatible future release could alter control semantics or execute arbitrary Python during import. The audit found no evidence that the currently intended `techmanpy` project is malicious. The issue is the unresolved and unverified dependency policy. ### Attack Path 1. A user installs dependencies according to the manifest or documentation at a later date. 2. The package resolver selects a newer release satisfying `>=1.0.0`, or installs an unnecessary third-party `asyncio` distribution. 3. The selected artifact contains compromised code, an incompatible API change, or behavior that was not part of the audited version. 4. The skill imports that package during robot connection or control. 5. The dependency executes with the user's process privileges and can affect both the host and robot-control traffic. Successful malicious-package exploitation depends on compromise or substitu ...[truncated 547 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (82)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The README documents live robot motion and stop commands (move_joints_zero, zero, stop) and I/O control capabilities without an explicit safety warning about physical movement, pinch/crush hazards, workspace clearance, E-stop readiness, or safe commissioning procedures. In a network-connected industrial robot context, users may run examples directly against production hardware, which can result in unintended motion, collisions, or actuation that may injure people or damage equipment.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · docs/svr_socket_content.txt (reported line 34)May include surrounding context.

text
Token 
Yes 
string 
API access token 
ActionTarget 
Yes 
int

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · docs/svr_socket_content.txt (reported line 381)May include surrounding context.

text
Token 
Yes 
string 
API access token 
ActionTarget 
Yes 
int

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · docs/svr_socket_content.txt (reported line 570)May include surrounding context.

text
Token 
Yes 
string 
API access token 
ActionTarget 
Yes 
int

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · docs/svr_socket_content.txt (reported line 758)May include surrounding context.

text
Token 
Yes 
string 
API access token 
ActionTarget 
Yes 
int

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · docs/svr_socket_content.txt (reported line 952)May include surrounding context.

text
Token 
Yes 
string 
API access token 
ActionTarget 
Yes 
int

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · docs/svr_socket_content.txt (reported line 1127)May include surrounding context.

text
Token 
Yes 
string 
API access token 
ActionTarget 
Yes 
int

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · docs/svr_socket_content.txt (reported line 1192)May include surrounding context.

text
Token 
Yes 
string 
API access token 
ActionTarget 
Yes 
int

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · docs/svr_socket_content.txt (reported line 1260)May include surrounding context.

text
Token 
Yes 
string 
API access token 
ActionTarget 
Yes 
int

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · docs/variable_system_content.txt (reported line 446)May include surrounding context.

text
projects. To run the Warp node at its best efficiency, mind each of the following as listed: 
 
1. 
No warning while the project is running. 
2. 
Do not insert the dongle key. 
3.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

This example issues real robot motion commands immediately without any explicit safety interlock, workspace check, or user confirmation. In an industrial robotics context, executing movement from sample code can cause collision, equipment damage, or operator injury if run against a live robot.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The cartesian motion example modifies the current pose and moves the robot without warning the user or validating that the vertical path is clear. Even a simple Z-axis offset can be hazardous on a live manipulator if tooling, fixtures, or people are present in the workspace.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The test runner automatically calls movement routines as part of a general 'run all tests' flow, which increases the chance that a user executes live robot motion unintentionally. Auto-running physical actuation from example/test code is especially dangerous because users may expect tests to be non-destructive and safe.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

This demo initiates real robot motion commands, including homing and movement to a target joint configuration, without any explicit safety interlock, simulation-only mode, operator acknowledgment, or warning that physical movement will occur. In a robotics context, executing movement from an example script can cause unexpected motion, collision, or injury if a user runs it on connected hardware assuming it is a harmless demo.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The full demo combines connection, monitoring, robot motion, and digital output control in one workflow, increasing the chance that a user runs it end-to-end on a live system without appreciating that it performs hazardous physical actions. Combining multiple actuator-affecting behaviors without a safety gate materially raises operational risk, especially in industrial robotics where motion and output changes can interact with tooling and nearby equipment.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script executes reset_alarm() immediately after the stop routine with no user confirmation, hazard check, or verification of why the alarm was raised. On a physical robot, clearing alarms without operator review can re-enable motion readiness or mask unsafe states, creating a real safety risk if faults are bypassed during testing.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This script connects to a physical robot and immediately issues movement commands, including joint, linear, circular, and homing motions, without any operator acknowledgment, safety interlock, dry-run mode, or environment check. In a robotics context, automatic motion can cause collisions, equipment damage, or human injury if the workspace is not cleared or the robot state is unexpected.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The script directly invokes a physical robot arc motion command as soon as the code reaches that branch, with no operator confirmation, interlock check, simulation mode, or workspace safety validation. In a robotics context this is dangerous because unexpected motion can cause collision, equipment damage, or human injury if the robot is connected to a live system.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The linear move command is issued immediately to a hard-coded target pose without any confirmation or runtime safety validation. Because this controls real hardware over the network, an unintended run could move the robot into fixtures, people, or unreachable/collision-prone positions.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

Even the 'return to safe position' joint move is still an immediate live motion command and assumes the named pose is actually safe in the current environment. If the robot's current tooling, fixtures, or surroundings differ from expectations, this motion can itself be hazardous.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The script initiates physical robot motion automatically via asyncio.run(test()) and calls move_joint(...) without any operator confirmation, interlock check, or pre-actuation safety gate. In a robotics context, unsolicited motion is inherently dangerous because it can cause collisions, equipment damage, or human injury if the workspace is occupied or the robot state is not as expected.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

This script issues real robot motion commands directly after connecting and reading pose, with no explicit operator confirmation, interlock check, simulation/dry-run mode, workspace validation, or emergency-stop readiness prompt. In a robotics context, moving +100mm on X based solely on current pose can cause collisions, fixture damage, or injury if the robot is not in a verified safe state.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The joint motion interface allows immediate physical robot movement with no confirmation, interlock check, workspace validation, or user-facing safety warning. In a collaborative robot setting, exposing direct motion primitives without pre-execution safeguards can lead to unintended motion, collisions, equipment damage, or operator injury if invoked by mistake or through higher-level automation.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The linear, absolute, relative, and circular motion methods perform safety-critical path motion without validating target coordinates, confirming operator intent, or checking whether the current robot state permits safe execution. This is especially dangerous because path-based motion can sweep through occupied space, so even a valid endpoint may still create collisions or human-safety hazards.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The entire checklist and release guidance are written in Chinese, and there is no indication that users may choose another language or that the skill is intentionally limited to a Chinese-only audience. Under the stated policy, forcing a specific language without opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.