Back to skill

Security audit

glic-check

Security checks for vulnerabilities and agentic risk

Overview

This is a structured review skill that reads user-scoped files and reports quality findings without hidden persistence, credential use, network execution, or automatic fixes.

Install this if you want an agent to perform structured quality reviews over files you name. Use explicit targets because the skill is designed to read files in scope, and review any suggested fixes before approving them.

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
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (8)

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · references/dimensions.md (reported line 173)May include surrounding context.

md
- [ ] File system side effects: writes to unexpected locations? overwrites existing files without warning?

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 · references/examples.md (reported line 105)May include surrounding context.

md
❌ **I1: SKILL.md claims "file upload supported" but `send.py` does not handle empty files**
`send.py:892` does not check file size; an empty file causes an API error.

⚠️ **I2: `credentials.json` is in `.gitignore` but SKILL.md does not say how to obtain it**
New users following SKILL.md will hit an auth failure.

### C — Containment

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill metadata advertises broad, everyday phrases such as "systematically review this," "audit my skill," and "quality check this code," which can match many ordinary user requests outside the intended narrow invocation pattern. That increases the chance of unintended activation, causing the skill to run in contexts where the user did not explicitly request this behavior and potentially altering agent behavior or exposing unrelated content to a review workflow.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The mode trigger table includes vague examples like "glic this" and "uglic this," which are weakly constrained and may overlap with casual conversation or unrelated references. In a skill-routing system, ambiguous activation examples can lead to false-positive routing and execution of this skill when a different action was intended.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% 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 · references/dimensions.md (reported line 178)May include surrounding context.

md
- [ ] Security: credential exposure (hardcoded keys, tokens in logs), path traversal, injection risks
- [ ] Backward compatibility: existing callers/importers unaffected; default behavior unchanged
- [ ] Scope match: does the change do more than what the commit message / PR description says?
- [ ] Input validation: user/external input sanitized; no trust without verification

### skill
- [ ] Token impact: is new content essential? does it justify its context window cost?

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% 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 · references/output-format.md (reported line 145)May include surrounding context.

Want me to fix these?

text

For ERR items, fix is expected. For WARN items, offer but don't push. Never auto-fix without confirmation.

## Proportions

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

Line L029 hard-codes specific Chinese phrases as examples of prohibited vague directives, and multiple section headers use Chinese labels (for example, 'Grammar(语法/格式/命名)'). In a user-facing document, this imposes a specific language/locale without stating that multilingual content is optional or justified for a region-specific audience.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

This markdown file includes the phrase "copy my profile" as an example of a trigger being too broad. In a markdown file, broad invocation wording that overlaps with ordinary user speech can cause unintended activation if reused as guidance.

Content

No source excerpt is available for this finding.