ClawHub
Back to skill

Security audit

Pilot Workflow

Security checks for vulnerabilities and agentic risk

Overview

The skill is coherent for multi-agent workflow orchestration, but it can send workflow tasks to dynamically selected agents without clear trust, scoping, or confirmation guidance.

Review workflows before use and assume task text, URLs, identifiers, and results may be sent to Pilot Protocol peers. Prefer explicit agent addresses or trusted allowlists over tag-based selection, and avoid putting secrets or private data in workflow tasks unless the destination agents are verified.

A.I.G

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (5)

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The prose at L038 states that the skill enables conditional branching, loops, parallel execution, and event-driven triggers. However, the actual engine example at L100-L138 simply iterates linearly over .steps, submits each task, and waits for completion, without evaluating condition, handling loops, running steps in parallel, or acting on triggers. This is an active contradiction between the documented capability and the code shown.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 55)May include surrounding context.

md
steps:
  - id: fetch
    agent: tag:api-gateway
    task: "Fetch data from https://api.example.com/data"

  - id: validate
    depends_on: fetch

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The workflow example instructs an external agent to fetch data from a remote URL and does so without any user-facing warning that data and task contents may be transmitted off-host. In an orchestration skill, this can lead users to unknowingly expose sensitive prompts, identifiers, or internal workflow context to remote services or third-party agents.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file presents a YAML schema using depends_on and condition at L57-L66, then labels the Bash script at L84-L141 as a 'Complete workflow engine'. In reality, the script never reads depends_on, condition, or triggers; it only fetches id, agent, and task before executing every step in order. Calling this engine 'complete' contradicts what the code actually supports.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documented engine submits tasks to discovered agents and monitors them through pilotctl without clearly warning that workflow content is being sent to external peers. Because agent selection is dynamic and based on tags, users may unknowingly dispatch sensitive instructions or data to untrusted or unintended remote systems.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.