Back to skill

Security audit

huawei-cloud-cloudrobo-robot

Security checks for vulnerabilities and agentic risk

Overview

This skill needs review because its tests can change or delete robot registrations without confirmation, and its setup installs unverified telemetry software.

Avoid running the bundled tests against existing robots. Require explicit approval for each mutation and restrict lifecycle tests to disposable registrations. Make telemetry installation opt-in and verify a pinned release before execution; SKILL_QUALITY_REPORT=0 is documented as disabling reporting, not installation. Use unique certificate passwords and protected export directories. The downloaded telemetry code was unavailable for inspection, so its data handling remains uncertain.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/test-cli-commands.sh
Finding

Lifecycle verification submits robot deletion without affirmative confirmation

Content
View full analysis

Impact

Unattended verification can submit deletion of an existing robot registration and overwrite its description without affirmative approval. Successful deletion removes the registration; it does not establish physical destruction of the robot.

Vulnerability Patterns
  • 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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
Findings (21)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

This finding describes the same auxiliary-tool installation risk as the preceding finding. Installing a persistent telemetry tool is not necessary to manage robots, and exposes the credential-bearing agent environment to additional executable code.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill makes automatic installation of an auxiliary telemetry executable part of robot management, extending execution and persistent filesystem changes beyond the requested cloud operations. The dependency is disclosed, so this is not evidence of concealment, but its automatic installation introduces an unnecessary supply-chain trust boundary in an environment handling cloud credentials.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Most commands directly match the description. However, Test 11 explicitly performs a cross-module workspace-listing operation. This accesses a separate resource category and is not needed to scope robot commands to the supplied workspace ID, so it constitutes an undeclared capability rather than a supporting implementation detail.

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 · SKILL.md (reported line 159)May include surrounding context.

md
> **"导出配置文件" = 导出接入配置(证书)**. A robot has **two** configurations:
> - **本体配置(robot body/main config)** — built into `r2c_sdk`, or distributed together with the
>   robot adapter alongside other robots. It is **not** what "download/export config" refers to.
> - **接入配置(access config)** — the robot's access credential bundle used to connect to the
>   platform. This is the one you download/export, and it is produced by `export-certificate`.
>
> The downloaded access config is a **zip** package — treat it as a sensitive credential bundle and

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · SKILL.md (reported line 214)May include surrounding context.

md
record the `service_id` / `exec_model_id`.
4. Pass the `robot_id` and model to `robo-dispatcher`: `cloudrobo dispatch create-task --session-id <sid> --name <name> --task "<task>" --constraints-json '{"model":{"exec_model_id":"<exec_model_id>"},"robot_id":"<robot_id>","exec_constraints":{"max_run_time":10,"max_iter_num":100}}'`.
   > This skill does not call cloudrobo-infer/dispatch by name; the agent orchestrates across
   > skills by first obtaining the robot_id here, then using the infer and dispatch skills.

### Cleanup Workflow (module)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · references/iam-policies.md (reported line 59)May include surrounding context.

md
| Create robot | Write access to `cloudrobo-service` (`POST /v1/robots`) |
| Show robot detail | Read access to `cloudrobo-service` (`GET /v1/robots/{robot_id}`) |
| Update robot | Write access to `cloudrobo-service` (`PUT /v1/robots/{robot_id}`) |
| Delete robot | Delete access to `cloudrobo-service` (`DELETE /v1/robots/{robot_id}`) |
| Export certificate | Write access to `cloudrobo-service` (`POST /v1/robots/{robot_id}/certificate/export`) |
| Query SDK info | Read access to `cloudrobo-service` (`GET /v1/robots/sdk`) |

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · references/robot-config-catalog.md (reported line 90)May include surrounding context.

md
| list_robots | `list_robots(**params)` | `list` | `GET /v1/robots` |
| show_robot | `show_robot(robot_id)` | `show` | `GET /v1/robots/{robot_id}` |
| update_robot | `update_robot(robot_id, req)` | `update` | `PUT /v1/robots/{robot_id}` |
| delete_robot | `delete_robot(robot_id)` | `delete` | `DELETE /v1/robots/{robot_id}` |
| export_robot_certificate | `export_robot_certificate(robot_id, req)` | `export-certificate` | `POST /v1/robots/{robot_id}/certificate/export` |
| show_sdk | `show_sdk()` | `show-sdk` | `GET /v1/robots/sdk` |

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

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

Content

Scanner excerpt · references/verification-method.md (reported line 87)May include surrounding context.

md
# → Expected click usage error: "not one of arm/humanoid/operation/other/quadruped/wheeled" (exit 2)

# 2. Path traversal rejected via validate_safe_id
cloudrobo robot show --robot-id "../etc/passwd"
# → Expected validation error

# 3. Missing workspace rejected

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

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

Content

Scanner excerpt · templates/test-vars.json (reported line 102)May include surrounding context.

json
{
      "id": "TC-13",
      "name": "path-traversal-rejected",
      "command": "cloudrobo robot show --robot-id '../etc/passwd'",
      "description": "Verify path traversal is rejected via validate_safe_id",
      "expected": "Error: validation error",
      "type": "read"

External Script Fetching

High
Category
Supply Chain
Confidence
94% confidence
Finding

The matched pipe parses JSON with fixed Python code; it does not directly execute the HTTP response. However, the returned version selects a subsequently downloaded executable archive that is installed without authenticity verification, creating a genuine remote-code supply-chain risk.

Content

Scanner excerpt · scripts/ensure_cli.sh (reported line 16)May include surrounding context.

sh
API_URL="https://skillsapi.developer.myhuaweicloud.com/api/quality/cli/latest"
OBS_BASE="https://obs-skills-repository.obs.cn-north-4.myhuaweicloud.com/skill-quality-cli"

V=$(curl -s "${API_URL}" | python3 -c 'import sys,json;print(json.load(sys.stdin)["version"])' 2>/dev/null)
if [ -z "$V" ]; then
    echo "警告: 无法获取 skill-quality-cli 最新版本,跳过安装" >&2
    exit 0

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · scripts/install_cli.sh (reported line 12)May include surrounding context.

sh
API_URL="https://skillsapi.developer.myhuaweicloud.com/api/quality/cli/latest"
OBS_BASE="https://obs-skills-repository.obs.cn-north-4.myhuaweicloud.com/skill-quality-cli"

V=$(curl -s "${API_URL}" | python3 -c 'import sys,json;print(json.load(sys.stdin)["version"])' 2>/dev/null)
if [ -z "$V" ]; then
    echo "无法获取 skill-quality-cli 最新版本" >&2
    exit 1

Tests labeled as requiring user confirmation execute mutations automatically after a timeout.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The tests describe mutations as requiring confirmation but execute them automatically after three seconds unless interrupted. In unattended or agent-driven execution, supplying an existing ROBOT_ID can cause metadata changes, certificate export, and attempted deletion of a real robot without affirmative authorization; successful deletion can disrupt its management and associated workflows.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
70% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · SKILL.md (reported line 190)May include surrounding context.

md
it.
3. **Store securely** — After export, place the downloaded zip under secure storage (a protected
   directory, secret manager, or the robot-side onboarding location). Do not leave it in a shared /
   world-readable path. This zip is consumed on the robot side (e.g. by `r2c_sdk` / the R2C client)
   together with the body config already carried inside the SDK or adapter.
4. **Feedback** — Prompt the user that the access-config zip was written to the specified file, and
   that it can be used for robot-side access configuration (e.g. with `r2c_sdk.cloudroboclient`).

Certificate-export workflow lacks a warning about sensitive access material and password exposure.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The executable example exports real access material using a fixed, publicly documented password and a directory created without explicit restrictive permissions. The password is also placed in command arguments, potentially exposing it through shell history or process inspection; users following this standalone test can leave recoverable credentials behind.

Content

No source excerpt is available for this finding.

Installs an unrelated skill-quality CLI persistently in the user's environment.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The script automatically downloads the latest auxiliary executable and copies it into a persistent user executable directory without verifying a signature or independently trusted digest. Compromise of the distribution service or artifact can therefore install attacker-controlled code with the user's privileges, potentially exposing cloud credentials when the tool is invoked.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% 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 · scripts/ensure_cli.sh (reported line 35)May include surrounding context.

sh
tar xzf "${TMPDIR}/sqc.tar.gz" -C "${TMPDIR}"

mkdir -p ~/.local/bin/skill-quality-cli.d
cp "${TMPDIR}/skill-quality-cli" ~/.local/bin/ 2>/dev/null
cp "${TMPDIR}/skill-quality-cli.bin" ~/.local/bin/ 2>/dev/null
cp "${TMPDIR}/skill-quality-cli.d/cli_entry.py" ~/.local/bin/skill-quality-cli.d/ 2>/dev/null

Installs an unrelated skill-quality CLI outside the declared robot-management purpose.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The installer downloads and persistently installs executable components for a quality CLI outside the declared robot-management functionality, without checking a trusted digest or signature. Compromise of the artifact distribution infrastructure could therefore lead to arbitrary code execution when the installed CLI is subsequently invoked; manual installation and HTTPS reduce exposure but do not establish artifact integrity.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% 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 · scripts/install_cli.sh (reported line 25)May include surrounding context.

sh
curl -fsSL -o "${TMPDIR}/sqc.tar.gz" "${OBS_BASE}/v${V}/skill-quality-cli-v${V}-linux-${ARCH}.tar.gz"
tar xzf "${TMPDIR}/sqc.tar.gz" -C "${TMPDIR}"

mkdir -p ~/.local/bin/skill-quality-cli.d
cp "${TMPDIR}/skill-quality-cli" ~/.local/bin/ 2>/dev/null
cp "${TMPDIR}/skill-quality-cli.bin" ~/.local/bin/ 2>/dev/null
cp "${TMPDIR}/skill-quality-cli.d/cli_entry.py" ~/.local/bin/skill-quality-cli.d/ 2>/dev/null

The robot-only installation command installs the documented aggregate package instead.

Low
Category
Excessive Agency
Confidence
97% confidence
Finding

The section promises to install only the robot package, but L016 runs the same aggregate-package installation that L008–L011 explicitly describe as installing all functional modules. This contradicts the stated limited installation intent and installs unrelated modules rather than the advertised robot-only setup.

Content

No source excerpt is available for this finding.

The robot test script enumerates workspaces outside its stated robot-management purpose.

Low
Category
Excessive Agency
Confidence
92% confidence
Finding

When a workspace ID is supplied, the script invokes the separate workspace module to list workspaces rather than limiting its queries to robots in that workspace. The manifest describes robot lifecycle, certificate, and SDK operations, not general workspace enumeration; this cross-module listing is not needed to use the supplied workspace ID.

Content

No source excerpt is available for this finding.