Back to skill

Security audit

Vigil

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent safety-check wrapper, but users should pin and sandbox its external npm dependency before relying on it.

Before installing, pin vigil-agent-safety to a reviewed exact version, avoid unpinned npx execution, and run the checker with minimal filesystem, network, and environment-variable access. Do not pass secrets in tool parameters unless they are required for policy evaluation.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:17
Finding

Unpinned Third-Party Package Installation and Execution

Content
View full analysis
Remediation
View remediation
``` 2. Commit a `package.json` and `package-lock.json` containing the exact resolved version and npm integrity hash. 3. Replace implicit `npx` retrieval with execution of the pinned local binary: ```bash npm exec --offline -- vigil-agent-safety check --tool exec --params '{"command":"ls -la"}' ``` Alternatively, invoke `./node_modules/.bin/vigil-agent-safety` after a verified installation. 4. Audit or vendor the source code corresponding to the pinned package release so that the implementation included in the security boundary can be reviewed. 5. Install dependencies with lifecycle scripts disabled where compatible: ```bash npm ci --ignore-scripts ``` 6. Run the checker in a restricted environment with: - No unnecessary write access. - No access to SSH keys, cloud credentials, or unrelated secrets. - Denied outbound network access unless explicitly required. - A minimal environment-variable allowlist. - Child-process restrictions where supported. 7. Avoid passing secrets in tool parameters when they are not necessary for policy evaluation. Redact tokens, credentials, and sensitive request data before calling `checkAction`. 8. Add automated dependency review, integrity verification, and vulnerability monitoring for the pinned release. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (13)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL.md (reported line 30)May include surrounding context.

md
const result = checkAction({
  agent: 'my-agent',
  tool: 'exec',
  params: { command: 'rm -rf /' },
});

// result.decision === "BLOCK"

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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 · SKILL.md (reported line 72)May include surrounding context.

md
const result = checkAction({
  agent: 'my-agent',
  tool: 'exec',
  params: { command: 'rm -rf /' },
});

// result.decision === "BLOCK"

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% 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 · SKILL.md (reported line 30)May include surrounding context.

md
const result = checkAction({
  agent: 'my-agent',
  tool: 'exec',
  params: { command: 'rm -rf /' },
});

// result.decision === "BLOCK"

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

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

Content

Scanner excerpt · SKILL.md (reported line 42)May include surrounding context.

md
- Destructive commands (rm -rf, mkfs, reverse shells) → BLOCK
- SSRF (metadata endpoints, localhost, internal IPs) → BLOCK
- Data exfiltration (curl to external, .ssh/id_rsa access) → BLOCK
- SQL injection (DROP TABLE, UNION SELECT) → BLOCK
- Path traversal (../../../etc/shadow) → BLOCK
- Prompt injection (ignore instructions, [INST] tags) → BLOCK

Credential Access

High
Category
Privilege Escalation
Confidence
95% confidence
Finding

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

Content

Scanner excerpt · SKILL.md (reported line 44)May include surrounding context.

md
- SSRF (metadata endpoints, localhost, internal IPs) → BLOCK
- Data exfiltration (curl to external, .ssh/id_rsa access) → BLOCK
- SQL injection (DROP TABLE, UNION SELECT) → BLOCK
- Path traversal (../../../etc/shadow) → BLOCK
- Prompt injection (ignore instructions, [INST] tags) → BLOCK
- Encoding attacks (base64 decode, eval(atob())) → BLOCK
- Credential leaks (API keys, AWS keys, tokens) → ESCALATE

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% 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 · SKILL.md (reported line 72)May include surrounding context.

bash
# Check a tool call
node scripts/vigil-check.js exec '{"command":"rm -rf /"}'
# → BLOCK: Destructive command pattern

# Check a safe call

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
100% confidence
Finding

This duplicate match points to the same documented destructive command example, which remains dangerous because it provides a copy-paste payload for tool abuse. In an agent skill, such examples are especially risky because they can be treated as canonical usage patterns by integrators or automated systems.

Content

Scanner excerpt · scripts/vigil-check.js (reported line 7)May include surrounding context.

js
* Usage: node vigil-check.js <tool> <params_json>
 * 
 * Examples:
 *   node vigil-check.js exec '{"command":"rm -rf /"}'
 *   node vigil-check.js read '{"path":"../../../etc/passwd"}'
 *   node vigil-check.js web_fetch '{"url":"http://169.254.169.254/latest/meta-data"}'
 */

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
95% confidence
Finding

This duplicate match points to the same documented destructive command example, which remains dangerous because it provides a copy-paste payload for tool abuse. In an agent skill, such examples are especially risky because they can be treated as canonical usage patterns by integrators or automated systems.

Content

Scanner excerpt · scripts/vigil-check.js (reported line 7)May include surrounding context.

js
* Usage: node vigil-check.js <tool> <params_json>
 * 
 * Examples:
 *   node vigil-check.js exec '{"command":"rm -rf /"}'
 *   node vigil-check.js read '{"path":"../../../etc/passwd"}'
 *   node vigil-check.js web_fetch '{"url":"http://169.254.169.254/latest/meta-data"}'
 */

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

The documentation includes a concrete example for reading ../../../etc/passwd, which is a classic sensitive-file access and path traversal target. In a skill wrapper that accepts arbitrary tool and parameter input, this example materially increases the risk of credential or system-information targeting by demonstrating the exact payload shape expected by the interface.

Content

Scanner excerpt · scripts/vigil-check.js (reported line 8)May include surrounding context.

js
* 
 * Examples:
 *   node vigil-check.js exec '{"command":"rm -rf /"}'
 *   node vigil-check.js read '{"path":"../../../etc/passwd"}'
 *   node vigil-check.js web_fetch '{"url":"http://169.254.169.254/latest/meta-data"}'
 */

Cloud Metadata Access

High
Category
Server-Side Request Forgery
Confidence
94% confidence
Finding

The example URL targets 169.254.169.254, the well-known cloud metadata endpoint, which is commonly exploited via SSRF to steal instance credentials and environment details. In the context of an agent-integrated tool wrapper, documenting this exact input can facilitate misuse and signals that requests to sensitive internal addresses are within the expected parameter shape.

Content

Scanner excerpt · scripts/vigil-check.js (reported line 9)May include surrounding context.

js
* Examples:
 *   node vigil-check.js exec '{"command":"rm -rf /"}'
 *   node vigil-check.js read '{"path":"../../../etc/passwd"}'
 *   node vigil-check.js web_fetch '{"url":"http://169.254.169.254/latest/meta-data"}'
 */

async function main() {

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The file is positioned as a safety-check wrapper, but its documentation advertises obviously dangerous inputs such as destructive shell commands, path traversal to sensitive files, and cloud metadata access. In an agent-skill context, such examples normalize misuse and can be copied directly by users or downstream tooling, increasing the chance the wrapper is used to validate harmful actions rather than discourage them.

Content

No source excerpt is available for this finding.