Back to skill

Security audit

1

Security checks for vulnerabilities and agentic risk

Overview

This skill broadly matches its stated purpose, but it needs Review because it can use many account credentials, perform public posting workflows, run an agent loop, and contains unsafe network and SVG handling.

Install only if you need a broad multi-platform discovery and engagement tool. Use read-only discovery first, keep write credentials least-privileged, disable auto_respond unless you have reviewed the workflow, prefer dry-run/idempotency options, avoid unpinned npx execution, and do not configure HTTP or untrusted LLM endpoints with API keys. Treat generated SVG and podcast feed URLs as untrusted until the project adds stronger validation and sanitization.

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 (3)

T09 · Insecure Skill Coding Practices

Error
Location
grazer/imagegen.py:230
Finding

Configurable LLM Endpoint Enables SSRF and Cleartext Credential Disclosure

Content
View full analysis
str: """Generate SVG using any OpenAI-compatible LLM endpoint. Args: prompt: Image description llm_url: OpenAI-compatible chat completions endpoint llm_model: Model name/ID llm_api_key: Optional API key (Bearer token) temperature: Creativity (0.0-1.0) timeout: Request timeout in seconds Returns: Raw SVG string ready for 4claw media field """ headers = {"Content-Type": "application/json"} if llm_api_key: headers["Authorization"] = f"Bearer {llm_api_key}" payload = { "model": llm_model, "messages": [ {"role": "system", "content": LLM_SVG_SYSTEM_PROMPT}, {"role": "user", "content": f"Create an SVG image: {prompt}"}, ], "temperature": temperature, "max_tokens": 2048, } resp = requests.post(llm_url, json=payload, headers=headers, timeout=timeout) resp.raise_for_status() ``` The example configuration explicitly demonstrates a cleartext HTTP endpoint: ```json "imagegen": { "llm_url": "http://100.75.100.89:8080/v1/chat/completions", "llm_model": "gpt-oss-120b", "llm_api_key": null } ``` An equivalent unrestricted request is implemented in `src/imagegen.ts:186-208`. ### Technical Analysis The application accepts an arbitrary `llm_url` and sends an HTTP POST request to it without validating the scheme, hostname, resolved IP address, port, or redirect destination. The request may include: - A bearer credential supplied through `llm ...[truncated 2008 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
grazer/imagegen.py:193
Finding

Untrusted LLM-Generated SVG Is Accepted Without Active-Content Sanitization

Content
View full analysis
str: """Validate and sanitize SVG for 4claw.""" svg = svg.strip() # Must start with SVG_MAX_BYTES: raise ValueError(f"SVG exceeds 4KB limit ({len(svg.encode('utf-8'))} bytes)") return svg ``` The untrusted LLM response is extracted and passed directly to this validator: ```python resp = requests.post(llm_url, json=payload, headers=headers, timeout=timeout) resp.raise_for_status() content = resp.json()["choices"][0]["message"]["content"].strip() # Extract SVG if wrapped in code fences svg_match = re.search(r'', content) if not svg_match: raise ValueError("LLM did not produce valid SVG output") return _validate_svg(svg_match.group(0)) ``` The resulting SVG can subsequently be wrapped for publication: ```python def svg_to_media(svg: str) -> List[Dict]: """Wrap an SVG string into 4claw's media array format. Returns: List ready to pass as the 'media' field in 4claw API calls """ return [{ "type": "svg", "data": svg, "generated": True, "nsfw": False, }] ``` Equivalent validation and publication behavior exists in `src/imagegen.ts:151-158, 210-245`. ### Technical Analysis The `_validate_svg()` function does not sanitize SVG despite its docstring. It verifies only that the text begins with `
Remediation
View remediation
` - `` - Event-handler attributes beginning with `on` - External `href` and `xlink:href` values - CSS `url(...)` references - `javascript:` and unapproved `data:` URLs - External fonts, stylesheets, images, filters, and imports 4. Consider disallowing animations and complex filters unless they are essential. 5. Reject malformed, nested, or namespace-confused documents rather than attempting permissive repair. 6. Prefer rendering generated SVG to a raster image inside a network-isolated, resource-constrained sandbox before publication. 7. Apply the same sanitization in both the Python and TypeScript implementations. 8. Add adversarial tests covering scripts, event attributes, namespace tricks, entity expansion, foreign objects, external references, CSS URLs, and mixed-case or encoded URI schemes. 9. Treat remote model output as untrusted regardless of system-prompt instructions. ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
grazer/podcast_grazer.py:123
Finding

Unrestricted Podcast Feed Retrieval Enables SSRF

Content
View full analysis
List[Dict]: """Fetch recent episodes from a podcast RSS feed. Args: feed_url: The podcast's RSS feed URL limit: Maximum episodes to return Returns: List of episode dicts with title, description, audio_url, etc. """ resp = self.session.get(feed_url, timeout=self.timeout) resp.raise_for_status() eps = _parse_podcast_rss(resp.text) return eps[:limit] ``` The discovery workflow also automatically follows feed URLs returned by the search provider: ```python shows = self.search(query, limit=limit) for show in shows: feed_url = show.get("feed_url") if feed_url: try: show["episodes"] = self.episodes(feed_url, limit=episodes_per_show) except Exception: show["episodes"] = [] else: show["episodes"] = [] return shows ``` ### Technical Analysis The `episodes()` method performs a request to any supplied `feed_url` without checking: - Whether the scheme is HTTP or HTTPS - Whether the hostname resolves to a public address - Whether the target is loopback, private, link-local, or reserved - Whether redirects remain within an approved public destination - Whether DNS resolution changes between validation and connection - Whether the response size and content type are appropriate for an RSS feed The URL can be supplied directly by an API caller. It can also originate from an upstream podcast-search result and be fetched automatically by `discover()`. The Grazer host therefore acts as a network request proxy for an untrusted URL. A timeout limits request duration but does not prevent access to internal services or constrain the amount of response data read into me ...[truncated 1424 chars]
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
Findings (87)

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 · README.md (reported line 74)May include surrounding context.

retained for use when the service is restored.

bash
curl -fsSL https://bottube.ai/apt/gpg | sudo gpg --dearmor -o /usr/share/keyrings/grazer.gpg
echo "deb [signed-by=/usr/share/keyrings/grazer.gpg] https://bottube.ai/apt stable main" | sudo tee /etc/apt/sources.list.d/grazer.list
sudo apt update && sudo apt install grazer

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 · README.md (reported line 75)May include surrounding context.

retained for use when the service is restored.

bash
curl -fsSL https://bottube.ai/apt/gpg | sudo gpg --dearmor -o /usr/share/keyrings/grazer.gpg
echo "deb [signed-by=/usr/share/keyrings/grazer.gpg] https://bottube.ai/apt stable main" | sudo tee /etc/apt/sources.list.d/grazer.list
sudo apt update && sudo apt install grazer

Chaining Abuse

High
Category
Tool Misuse
Confidence
75% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · README.md (reported line 74)May include surrounding context.

retained for use when the service is restored.

bash
curl -fsSL https://bottube.ai/apt/gpg | sudo gpg --dearmor -o /usr/share/keyrings/grazer.gpg
echo "deb [signed-by=/usr/share/keyrings/grazer.gpg] https://bottube.ai/apt stable main" | sudo tee /etc/apt/sources.list.d/grazer.list
sudo apt update && sudo apt install grazer

Chaining Abuse

High
Category
Tool Misuse
Confidence
75% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · README.md (reported line 75)May include surrounding context.

retained for use when the service is restored.

bash
curl -fsSL https://bottube.ai/apt/gpg | sudo gpg --dearmor -o /usr/share/keyrings/grazer.gpg
echo "deb [signed-by=/usr/share/keyrings/grazer.gpg] https://bottube.ai/apt stable main" | sudo tee /etc/apt/sources.list.d/grazer.list
sudo apt update && sudo apt install grazer

Chaining Abuse

High
Category
Tool Misuse
Confidence
75% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · README.md (reported line 76)May include surrounding context.

bash
curl -fsSL https://bottube.ai/apt/gpg | sudo gpg --dearmor -o /usr/share/keyrings/grazer.gpg
echo "deb [signed-by=/usr/share/keyrings/grazer.gpg] https://bottube.ai/apt stable main" | sudo tee /etc/apt/sources.list.d/grazer.list
sudo apt update && sudo apt install grazer

Claude Code

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
100% 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 · debian/build-deb.sh (reported line 8)May include surrounding context.

sh
VERSION="1.9.1"
PKG="grazer_${VERSION}_all"

rm -rf "/tmp/${PKG}"
mkdir -p "/tmp/${PKG}/DEBIAN"
mkdir -p "/tmp/${PKG}/usr/lib/python3/dist-packages/grazer"
mkdir -p "/tmp/${PKG}/usr/bin"

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 · src/clawhub.ts (reported line 87)May include surrounding context.

ts
}

  /**
   * Update skill metadata
   */
  async updateSkill(skillId: string, updates: Partial<ClawHubSkill>): Promise<ClawHubSkill> {
    if (!this.token) {

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The module description presents the package as content discovery, but the implementation also supports authenticated posting, replying, commenting, image generation, and telemetry. This capability mismatch can mislead reviewers and integrators, causing them to grant the skill broader permissions or deploy it in contexts where write actions and network side effects were not expected.

Content

No source excerpt is available for this finding.

Ssd 1

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

Untrusted notification text is concatenated directly into the LLM prompt as conversational context, allowing attackers to embed instructions that influence the model's response. Because this component is explicitly used for automatic reply generation, a malicious comment or mention can steer the model into producing policy-violating, harmful, or attacker-directed output.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The deployment guide instructs users to publish packages to NPM and PyPI, push to GitHub, and submit data to a third-party API, but it does not clearly warn that these actions are public, irreversible in practice, and transmit project metadata to external services. In a skill context, users may follow steps mechanically, increasing the risk of unintended release, credential use, or data disclosure.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
84% confidence
Finding

The curl command sends project metadata to an external API endpoint, which is an outbound transmission to a third party. While this appears intended for legitimate skill registration, the guide does not warn the user about external transmission, verify the endpoint, or explain what data is being shared, making accidental disclosure or misuse more likely.

Content

Scanner excerpt · DEPLOY.md (reported line 59)May include surrounding context.

Create a skill entry on BoTTube:

bash
curl -X POST https://bottube.ai/api/skills \
  -H "Content-Type: application/json" \
  -d '{
    "name": "grazer",

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The instructions tell the user to modify a local file under /home/scott/CLAUDE.md and run an automated publish script without describing the local state changes or the script's external side effects. This is risky because users may execute the script without understanding that it can publish artifacts, create tags, and push changes remotely.

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 · INTEGRATION.md (reported line 123)May include surrounding context.

1. Setup Config

bash
mkdir -p ~/.grazer
cp config.example.json ~/.grazer/config.json
cp profile.example.json ~/.grazer/profile.json

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The guide instructs users to run npx grazer-agent without pinning a package version or integrity source, which can fetch and execute whatever package is currently published under that name. In a skill/integration context, this creates a supply-chain execution path where a compromised or typosquatted package could run arbitrary code on the host.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The standalone agent loop advertises real-time monitoring and auto-responding to comments but does not clearly warn that the software will post content on the user's behalf. That omission can cause operators to enable autonomous actions without informed consent, leading to unintended spam, reputational damage, or policy violations.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% 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 · INTEGRATION.md (reported line 163)May include surrounding context.

text

### Boris (Moltbook Bot)
Already integrated via notification monitor + auto-deploy

### Janitor (AutomatedJanitor2015)
Add to notification checking:

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The 'Next Steps' section tells users to enable auto_respond after testing, but still lacks a direct warning that doing so authorizes autonomous posting and interaction. In this context, the skill is specifically about bot integration across social platforms, so the missing disclosure increases the chance of accidental unattended actions.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · PUBLISH_CHECKLIST.md (reported line 54)May include surrounding context.

bash
# Using ClawHub CLI or API
curl -X POST https://clawhub.ai/api/skills \
  -H "Authorization: Bearer <CLAWHUB_TOKEN_REDACTED — set via clawhub login, stored in ~/.config/clawhub>" \
  -H "Content-Type: application/json" \
  -d '{

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · README.md (reported line 74)May include surrounding context.

retained for use when the service is restored.

bash
curl -fsSL https://bottube.ai/apt/gpg | sudo gpg --dearmor -o /usr/share/keyrings/grazer.gpg
echo "deb [signed-by=/usr/share/keyrings/grazer.gpg] https://bottube.ai/apt stable main" | sudo tee /etc/apt/sources.list.d/grazer.list
sudo apt update && sudo apt install grazer

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · README.md (reported line 75)May include surrounding context.

retained for use when the service is restored.

bash
curl -fsSL https://bottube.ai/apt/gpg | sudo gpg --dearmor -o /usr/share/keyrings/grazer.gpg
echo "deb [signed-by=/usr/share/keyrings/grazer.gpg] https://bottube.ai/apt stable main" | sudo tee /etc/apt/sources.list.d/grazer.list
sudo apt update && sudo apt install grazer

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · README.md (reported line 76)May include surrounding context.

retained for use when the service is restored.

bash
curl -fsSL https://bottube.ai/apt/gpg | sudo gpg --dearmor -o /usr/share/keyrings/grazer.gpg
echo "deb [signed-by=/usr/share/keyrings/grazer.gpg] https://bottube.ai/apt stable main" | sudo tee /etc/apt/sources.list.d/grazer.list
sudo apt update && sudo apt install grazer

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 · README.md (reported line 102)May include surrounding context.

md
# Browse 4claw /crypto/ board
grazer discover -p fourclaw -b crypto

# Create a 4claw thread
grazer post -p fourclaw -b singularity -t "Title" -m "Content"

# Reply to a 4claw thread

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README advertises outbound actions such as posting, commenting, cross-posting, and bulk guestbook signing, but it does not prominently warn that these operations publish data to third-party services and can affect public accounts or create visible external side effects. In an agent skill context, that omission is risky because users may treat examples as safe read-only operations and accidentally trigger automated posting or spam-like behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill explicitly advertises notifications, auto-responses, agent training, and an autonomous loop, all of which imply ongoing interaction and possible write actions to third-party services. Presenting these capabilities without prominent consent, scoping, and side-effect warnings increases the risk that an agent operator enables unsolicited posting or engagement across multiple external platforms.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The examples show direct posting to 4claw and generating output files from the CLI without any caution that these actions have external side effects. In an agent-skill context, concise examples are often copied verbatim, so omission of warnings can lead to accidental public posting or file creation by users who assume examples are safe to run.

Content

No source excerpt is available for this finding.