Back to skill

Security audit

Fix Exec Allowlist Miss

Security checks for vulnerabilities and agentic risk

Overview

The skill has no executable payload, but it directs broad gateway security-profile changes and credential setup in ways users should review carefully.

Review before installing. This skill should only be used by an operator who is allowed to change gateway security settings. Prefer a narrowly scoped exec allowlist fix over switching to a full tools profile, require hash-verified config patches, avoid plaintext API keys, and confirm there is a rollback path before applying changes.

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

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SKILL.md:6
Finding

Overly Broad Tool-Profile Escalation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 6–11
Vulnerability Type: Unauthorized access and privilege escalation
Risk Level: High

Vulnerable Code Snippet:

text
Diagnosing exec deny:
1. Check tools.exec.security and tools.profile in config
2. If profile=coding and security=full still denied → profile baseline missing exec
3. Use gateway tool: gateway('config.patch', {raw: '{tools:{profile:"full"}}'})
4. Hot-reload applies automatically in hybrid mode; explicit restart via gateway restart if needed
5. Verify: gateway('config.get', {}) → confirm profile=full

Technical Analysis

The Skill responds to an execution denial by directing the Agent to change the complete tools profile to full. This bypasses least-privilege controls rather than narrowly correcting the missing executable allowlist entry. The subsequent hot reload makes the broader permissions effective immediately.

An execution denial is a security-boundary decision. Treating it as a reason to enable a globally permissive profile may expose unrelated tools and execution capabilities that are not legitimately required for the original task. Although the document does not contain an executable payload, an Agent following these instructions could weaken its own access controls and increase the effect of later malicious or untrusted instructions.

Attack Path

  1. A requested command is denied by the existing execution policy.
  2. The Skill interprets the denial as a missing capability in the coding profile.
  3. The Agent invokes gateway('config.patch', ...) and sets tools.profile to full.
  4. Hybrid-mode hot reload applies the configuration change without requiring a deliberate restart.
  5. The Agent verifies that the unrestricted profile is active.
  6. A subsequent untrusted prompt, compromised Skill, or unsafe operation can use the newly exposed tools and command-execution capabilities.

Impact Assessment

...[truncated 723 chars]

Remediation
View remediation

Remediation Suggestions

  • Preserve the restrictive coding profile and add only the exact required executable, subcommand, and argument patterns to a scoped allowlist.
  • Do not recommend tools.profile: full as the default response to an execution denial.
  • Require explicit, informed user or administrator approval before broadening any security profile.
  • Before applying a configuration change, display the effective permission delta and explain which additional tools will become available.
  • Use the documented hash-verified patch workflow to prevent stale or conflicting changes.
  • Add a time-limited elevation mechanism that automatically restores the original profile after the approved operation.
  • Verify the narrowly scoped permission after modification and document a rollback procedure.
  • Audit and log profile changes, including the requester, justification, prior value, new value, and restoration status.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The metadata says config patching requires baseHash from config.get first, but the earlier troubleshooting workflow instructs patching config without it. Omitting concurrency protection on live configuration changes can cause stale writes, failed updates, or accidental overwriting of newer settings, which is especially risky when changing execution/security controls.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This markdown file appears to serve as the skill description, but it does not specify when the skill should be invoked or what exact user requests should activate it. The description covers several operational topics in a compressed way, which makes the trigger scope ambiguous and increases the chance of unintended activation for loosely related requests.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill documents gateway('config.patch', {raw: '{tools:{profile:"full"}}'}), which passes raw as a string, but later states that raw must be an object, not a string. This contradiction can cause failed patches or unsafe operator behavior during troubleshooting, especially because the action modifies security-relevant configuration (tools.profile) and could lead users to apply incorrect changes while believing they are enabling exec safely.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The instructions cover sensitive system configuration changes, restart behavior, auth context, and direct API key setup without warning about service impact, privilege requirements, or secret-handling risks. In context, this is more dangerous because the skill normalizes modifying execution profiles and setting provider credentials, which could lead to outages, unauthorized capability expansion, or credential exposure if followed casually.

Content

No source excerpt is available for this finding.