Back to skill

Security audit

huawei-cloud-cloudrobo-dispatch

Security checks for vulnerabilities and agentic risk

Overview

This robot-control skill needs review because its verification script can trigger movement or attempt task cancellation without affirmative confirmation.

Review before installing or running verification. Make the helper read-only or dry-run by default, and require explicit approval of the workspace, robot, action, and task before live operations. Use an isolated test robot or simulator and narrowly scoped credentials. All nine bundled files were inspected; the external CLI and live service behavior remain outside this review. The evidence supports a control and disclosure concern, without establishing malicious intent.

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

Verification helper performs live robot operations without enforcing confirmation

Content
View full analysis

Impact

Verification can initiate unconfirmed robot movement or interrupt an existing robot task using the caller's authenticated workspace access.

Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (14)

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/dataflow-diagram.md (reported line 74)May include surrounding context.

md
A->>U: confirm cancellation (mutating op)
    U-->>A: confirm
    A->>CLI: cancel-task --session-id <sid> --task-id <tid>
    CLI->>S: DELETE /v1/robo-dispatcher/sessions/{sid}/tasks/{tid}
    S->>D: request cancellation
    D-->>S: task transitioned to terminal (cancelled)
    S-->>CLI: JSON

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 · references/dataflow-diagram.md (reported line 142)May include surrounding context.

md
> **Note**: This skill does not call other skills by name. The agent orchestrates across
> skills: it uses `cloudrobo-robot`/`cloudrobo-asset`/`cloudrobo-infer` first to resolve
> `robot_id`, model / `exec_model_id`, and `infer-service_id`, then uses this dispatch skill
> to create and monitor the dispatcher task.

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 · references/iam-policies.md (reported line 47)May include surrounding context.

md
- A task created in one session/workspace is invisible to another
- The robot and exec model referenced by a task must be valid within the same workspace

### Minimal Access for dispatch skill

| Capability | Required Permission |
| ----------- | ------------------- |

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 · scripts/test-cli-commands.sh (reported line 2)May include surrounding context.

sh
#!/bin/bash
# Test script for cloudrobo-dispatch skill
# Usage: ./test-cli-commands.sh
# Requires: a valid SESSION_ID, and for full lifecycle tests, a test ROBOT_ID and EXEC_MODEL_ID.
# Optional env vars:

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 · scripts/test-cli-commands.sh (reported line 20)May include surrounding context.

sh
TASK_ID=${TASK_ID:-}
INFER_SERVICE_ID=${INFER_SERVICE_ID:-}

echo "=== Testing cloudrobo-dispatch skill ==="
echo "Session ID: ${SESSION_ID:-<not set>}"
echo "Robot ID:   ${ROBOT_ID:-<not set>}"
echo "Exec Model: ${EXEC_MODEL_ID:-<not set>}"

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 55)May include surrounding context.

md
| List tasks | Read access to `cloudrobo-service` (`GET /v1/robo-dispatcher/sessions/{session_id}/tasks`) |
| Show task | Read access to `cloudrobo-service` (`GET /v1/robo-dispatcher/sessions/{session_id}/tasks/{task_id}`) |
| Wait for task (`wait-task`) | Same read access as show-task — `wait-task` is a client-side polling helper that repeatedly `GET`s the task (`GET /v1/robo-dispatcher/sessions/{session_id}/tasks/{task_id}`); no additional permission |
| Cancel task | Write access to `cloudrobo-service` (`DELETE /v1/robo-dispatcher/sessions/{session_id}/tasks/{task_id}`) |
| Show task result | Read access to `cloudrobo-service` (`GET /v1/robo-dispatcher/sessions/{session_id}/tasks/{task_id}/result`) |
| Resolve robot (cross) | Read access to `cloudrobo-service` robot list/show |
| Resolve exec model (cross) | Read access to `cloudrobo-asset-manager` / infer service |

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 62)May include surrounding context.

md
| Test | Command | Expected |
| ------ | ------- | ---------- |
| Path traversal | `cloudrobo dispatch show-task --session-id <sid> --task-id '../etc/passwd'` | Blocked by `validate_safe_id` — error raised |
| Missing session_id | `cloudrobo dispatch create-task` without `--session-id` | CLI requires the parameter; clear usage error |
| Invalid session_id | `cloudrobo dispatch list-tasks --session-id 'nonexistent'` | Appropriate not-found / validation error |
| Dry-run create | `cloudrobo dispatch create-task --... --dry-run` | Shows what would be created; no task submitted |

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 · scripts/test-cli-commands.sh (reported line 124)May include surrounding context.

sh
# Test 12: Path traversal blocked (exit-code based; no reliance on specific error text)
echo "Test 12: Path traversal blocked"
if TRAVERSAL_OUTPUT=$(cloudrobo dispatch show-task --session-id "$SESSION_ID" --task-id '../etc/passwd' 2>&1); then
    echo "FAIL: path traversal not blocked (command succeeded)"
else
    echo "PASS: traversal blocked (command exited non-zero; actual error below)"

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-blocked",
      "command": "cloudrobo dispatch show-task --session-id {session_id} --task-id '../etc/passwd'",
      "description": "Verify validate_safe_id blocks path traversal",
      "expected": "Error raised (invalid task id)",
      "type": "read"

Generic task-management trigger phrases can activate a robot-dispatch skill for unrelated requests.

Medium
Category
Trigger Abuse
Confidence
94% confidence
Finding

The trigger list includes unqualified phrases such as “dispatch,” “task scheduling,” “cancel task,” and “task result,” which also apply to ordinary software and workflow tasks. Although the description explains the CloudRobo domain, it does not explicitly require robot or CloudRobo context for these triggers or exclude unrelated task-management requests.

Content

No source excerpt is available for this finding.

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 29)May include surrounding context.

md
## Overview

The `cloudrobo-dispatch` skill manages **embodied task dispatch** via the `robo-dispatcher`
service. It lets the agent create a task that runs on a robot (identified by `robot_id`) under
an execution model (`exec_model_id`) inside a session (`session_id`), monitor it, cancel it, and
retrieve the natural-language task result plus log items.

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 · references/verification-method.md (reported line 67)May include surrounding context.

md
| Invalid session_id | `cloudrobo dispatch list-tasks --session-id 'nonexistent'` | Appropriate not-found / validation error |
| Dry-run create | `cloudrobo dispatch create-task --... --dry-run` | Shows what would be created; no task submitted |
| Dry-run cancel | `cloudrobo dispatch cancel-task --session-id <sid> --task-id <tid> --dry-run` | Shows what would be cancelled; no action taken |
| Confirm gate | create-task / cancel-task without confirmation | Agent prompts user before execution |

## Expected Results Matrix

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 · scripts/test-cli-commands.sh (reported line 143)May include surrounding context.

sh
echo "  (cancelling a finished task is rejected; see SKILL.md Edge Cases). Verify with show-task."
    fi
else
    echo "  (No TASK_ID; script did not verify cancellation capability. Create a task to obtain a TASK_ID and run cancel-task manually.)"
fi
echo ""

L150 describes the script as read-only, but it creates robot-motion tasks and cancels tasks.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

Despite describing its checks as read-only, the script can submit a real 'Move forward 1 meter' robot task and cancel an existing task supplied through TASK_ID. Creation uses only a two-second opt-out warning and cancellation has no affirmative confirmation gate, so running it as verification can cause unintended physical movement or interrupt ongoing operations.

Content

No source excerpt is available for this finding.