Back to skill

Security audit

Ghostprint

Security checks for vulnerabilities and agentic risk

Overview

Ghostprint openly sends realistic background LLM requests, but its credential handling, automatic scheduling, cron persistence, and unpinned install path need careful review before use.

Install only if you intentionally want automated synthetic LLM traffic from your accounts. Use separate low-privilege API keys or accounts for noise, pin the source to a reviewed commit, avoid custom base_url values unless credentials are endpoint-specific, review expected costs and provider policies, and do not use --install-cron unless you are prepared to inspect and remove the crontab entry manually.

Vulnerability Patterns
  • System PersistenceInstalls backdoors, hooks, services, or scheduled tasks that survive the run
  • 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
Findings (4)

T06 · System Persistence

Warning
Location
ghostprint.py:453
Finding

Persistent Execution Through User Crontab Modification

Content
View full analysis
> {shlex.quote(str(LOG_FILE))}\n" import subprocess result = subprocess.run(["crontab", "-l"], capture_output=True, text=True) existing = result.stdout if result.returncode == 0 else "" if str(script) in existing: print("ghostprint is already in crontab.") return new_cron = existing + cron_line proc = subprocess.run(["crontab", "-"], input=new_cron, text=True) ``` ### Technical Analysis The `--install-cron` operation reads the current user's crontab, appends a command that invokes `ghostprint.py`, and writes the modified crontab back. This causes Ghostprint to execute repeatedly across terminal sessions and system reboots where the cron service is active. Scheduled execution is relevant to the declared noise-generation functionality, and installation requires an explicit command. However, it still establishes operating-system-level persistence outside the OpenClaw lifecycle. The project does not provide a corresponding uninstall operation, uniquely identify the installed entry, or obtain an additional confirmation before modifying persistent user configuration. The configured interval is also interpolated directly into the cron expression without validating that it is a positive, cron-compatible integer. This is not a shell-injection path because the value is not interpreted by a shell during installation, but malformed values can create an invalid or unexpectedly frequent schedule. ### Attack Path 1. The user follows the standalone installation instructions. 2. The user invokes `python3 ghostprint.py --i ...[truncated 1047 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
openclaw-plugin/index.ts:452
Finding

Provider Credentials Can Be Transmitted to an Unrelated Configured Host

Content
View full analysis
=> { try { const auth = await api.runtime.modelAuth.resolveApiKeyForProvider({ provider, cfg: api.config, }); return auth?.apiKey ?? undefined; } catch { return undefined; } }; ``` The resolved credential is then sent to the separately configurable `base_url`: ```typescript async function fireOne(p: ProviderEntry, apiKey: string, opts: FireOptions): Promise { const base = p.base_url.replace(/\/$/, ""); const { prompt, maxTokens, temperature, systemPrompt } = opts; if (p.style === "anthropic") { const body: any = { model: p.model, max_tokens: maxTokens, temperature, messages: [{ role: "user", content: prompt }], }; if (systemPrompt) body.system = systemPrompt; const raw = await post(`${base}/messages`, { "x-api-key": apiKey, "anthropic-version": "2023-06-01", }, body); return JSON.parse(raw)?.content?.[0]?.text ?? "(no reply)"; } else { const messages: any[] = []; if (systemPrompt) messages.push({ role: "system", content: systemPrompt }); messages.push({ role: "user", content: prompt }); const raw = await post(`${base}/chat/completions`, { "Authorization": `Bearer ${apiKey}`, }, { model: p.model, max_tokens: maxTokens, temperature, messages }); const msg = JSON.parse(raw)?.choices?.[0]?.message; return msg?.content || msg?.reasoning_content || "(no reply)"; } } ``` ### Technical Analysis The plugin resolves an existing OpenClaw API key ...[truncated 2045 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
openclaw-plugin/index.ts:388
Finding

Hostname Validation Can Be Bypassed Through DNS Resolution or Rebinding

Content
View full analysis
= 16 && parts[1] <= 31) || (parts[0] === 192 && parts[1] === 168) || (parts[0] === 169 && parts[1] === 254) || parts[0] === 0 ) { throw new Error(`Rejected private/reserved IP in URL: ${urlStr}`); } } } function post(urlStr: string, headers: Record, body: object, proxyUrl?: string): Promise { return new Promise((resolve, reject) => { validateUrl(urlStr); const parsed = new URL(urlStr); const data = JSON.stringify(body); const opts = { hostname: parsed.hostname, port: Number(parsed.port) || 443, path: parsed.pathname + parsed.search, method: "POST", headers: { "Content-Type": "application/json", "Content-Length": Buffer.byteLength(data), ...headers, }, }; const req = https.request(opts, res => { let out = ""; res.on("data", chunk => (out += chunk)); res.on("end", () => res.statusCode && res.statusCode < 400 ? resolve(out) : reject(new Error(`HTTP ${res.statusCode}`)), ); }); req.on("error", reject); req.setTimeout( ...[truncated 3693 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
SKILL.md:12
Finding

Installation Instructions Execute an Unpinned Mutable Remote Source

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (36)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The documented behavior materially understates what the skill does: it reportedly installs or edits cron via subprocess, persists local logs/state, and requires separate provider configuration despite claims of seamless key reuse. This mismatch is dangerous because it defeats informed consent and can lead users to authorize a tool that establishes persistence, consumes paid API credits, and writes local artifacts without clear disclosure.

Content

No source excerpt is available for this finding.

Ssd 4

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The plugin narrative frames stealthy provider traffic as benign 'noise' while its actual purpose is to hide true usage patterns by blending in with synthetic activity. In skill context, this is more dangerous because it is packaged as an automation plugin that reuses existing provider credentials and runs on a schedule, enabling persistent deceptive network behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The plugin explicitly states that it is designed to be statistically indistinguishable from real human usage and to prevent fingerprinting by injecting realistic traffic. This is a stealth/evasion capability, not a normal productivity function, and it can be used to disguise actual usage patterns, consume third-party resources covertly, or defeat monitoring assumptions.

Content

No source excerpt is available for this finding.

Ssd 4

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The tool description explicitly advertises immediate generation of realistic-looking sessions to depersonalize usage fingerprints, confirming intent to automate deceptive imitation. Because this is exposed as an invocable tool, another agent or workflow could trigger it on demand to create covert-looking outbound traffic using the user’s configured API keys.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The tool can trigger outbound LLM traffic immediately and the plugin also schedules recurring background requests without explicit user confirmation at execution time. In context, the traffic is intentionally designed to mimic real usage and obscure behavioral fingerprints, so hidden or surprise transmission materially increases stealth and misuse risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest explicitly states that the plugin sends scheduled queries to external LLM providers, yet it provides no accompanying privacy warning about autonomous data transmission, API usage, retention by third parties, or billing implications. Because this skill is specifically designed to generate background traffic across multiple providers using configured keys, the lack of explicit warning materially increases the chance of silent privacy leakage, unexpected charges, and undisclosed outbound network activity.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This plain-text file explicitly says the goal is "organic-looking noise" and then enumerates many common everyday questions such as cooking, cleaning, and health queries. If these lines are used as activation topics or triggers, they heavily overlap with normal user speech and lack any scope constraints or exclusion conditions, creating a high risk of accidental invocation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This documentation promotes automatic transmission of synthetic prompts to third-party LLM providers using existing API credentials, but does not prominently warn users that this creates additional provider-side account activity, billing, logs, and possible policy or privacy consequences. In the context of an anti-fingerprinting tool, that omission is especially risky because users may incorrectly assume the feature is privacy-preserving by default when it actually increases data sharing and can strengthen account-level correlation if misconfigured.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README explicitly states that the plugin runs as a background service, automatically fires on a schedule, and reuses existing provider API keys. That creates ongoing third-party network activity and billable account usage that may not be obvious at install time, which is risky in an agent/plugin context where users may enable it once and forget it remains active.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill advertises automatic scheduling, provider credential reuse, and standalone execution, yet the manifest shown does not declare any tool scope or permissions despite requiring network, filesystem, environment, and shell-like capabilities. This is dangerous because users and host systems cannot accurately assess or constrain what the skill may access, increasing the chance of overprivileged execution and unexpected side effects such as background persistence or credential use.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill description does not clearly warn that it will automatically and repeatedly send randomized prompts to third-party LLM providers using the user's credentials. In this context, the omission is risky because the core function triggers ongoing external network activity, potential data exposure through prompts/metadata, and recurring cost, all of which should be explicitly surfaced before installation or enablement.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The standalone instructions tell users to install cron without clearly stating that this creates a persistent background task outside the current session. That is dangerous because users may unknowingly leave behind an automated process that continues making network requests, consuming API credits, and modifying system behavior after they believe they are done.

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 · README.md (reported line 72)May include surrounding context.

md
providers:
  - name: anthropic
    base_url: https://api.anthropic.com/v1
    api_key: ${ANTHROPIC_API_KEY}
    model: claude-haiku-4-5
    style: anthropic          # anthropic | openai

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 · README.md (reported line 128)May include surrounding context.

md
providers:
  - name: anthropic
    base_url: https://api.anthropic.com/v1
    api_key: ${ANTHROPIC_API_KEY}
    model: claude-haiku-4-5
    style: anthropic          # anthropic | openai

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 · config.example.yaml (reported line 7)May include surrounding context.

yaml
providers:
  - name: anthropic
    base_url: https://api.anthropic.com/v1
    api_key: ${ANTHROPIC_API_KEY}
    model: claude-haiku-4-5
    style: anthropic          # anthropic | openai

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 · openclaw-plugin/index.ts (reported line 302)May include surrounding context.

ts
providers:
  - name: anthropic
    base_url: https://api.anthropic.com/v1
    api_key: ${ANTHROPIC_API_KEY}
    model: claude-haiku-4-5
    style: anthropic          # anthropic | openai

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 · README.md (reported line 80)May include surrounding context.

md
weight: 1                 # used with strategy: weighted

  - name: zai
    base_url: https://api.z.ai/api/paas/v4
    api_key: ${ZAI_API_KEY}
    model: glm-4-flash
    style: openai

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 · README.md (reported line 135)May include surrounding context.

md
weight: 1                 # used with strategy: weighted

  - name: zai
    base_url: https://api.z.ai/api/paas/v4
    api_key: ${ZAI_API_KEY}
    model: glm-4-flash
    style: openai

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 · config.example.yaml (reported line 14)May include surrounding context.

yaml
weight: 1                 # used with strategy: weighted

  - name: zai
    base_url: https://api.z.ai/api/paas/v4
    api_key: ${ZAI_API_KEY}
    model: glm-4-flash
    style: openai

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 · openclaw-plugin/index.ts (reported line 310)May include surrounding context.

ts
weight: 1                 # used with strategy: weighted

  - name: zai
    base_url: https://api.z.ai/api/paas/v4
    api_key: ${ZAI_API_KEY}
    model: glm-4-flash
    style: openai

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 · README.md (reported line 157)May include surrounding context.

md
# Uncomment to add more:
  # - name: openai
  #   base_url: https://api.openai.com/v1
  #   api_key: ${OPENAI_API_KEY}
  #   model: gpt-4o-mini
  #   style: openai

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 · config.example.yaml (reported line 22)May include surrounding context.

yaml
# Uncomment to add more:
  # - name: openai
  #   base_url: https://api.openai.com/v1
  #   api_key: ${OPENAI_API_KEY}
  #   model: gpt-4o-mini
  #   style: openai

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 · README.md (reported line 164)May include surrounding context.

md
#   weight: 2

  # - name: mistral
  #   base_url: https://api.mistral.ai/v1
  #   api_key: ${MISTRAL_API_KEY}
  #   model: mistral-small-latest
  #   style: openai

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 · config.example.yaml (reported line 29)May include surrounding context.

yaml
#   weight: 2

  # - name: mistral
  #   base_url: https://api.mistral.ai/v1
  #   api_key: ${MISTRAL_API_KEY}
  #   model: mistral-small-latest
  #   style: openai

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 · config.example.yaml (reported line 35)May include surrounding context.

yaml
#   style: openai

  # - name: groq
  #   base_url: https://api.groq.com/openai/v1
  #   api_key: ${GROQ_API_KEY}
  #   model: llama3-8b-8192
  #   style: openai