Back to skill

Security audit

CGArtLab Person Profile

Security checks for vulnerabilities and agentic risk

Overview

This skill coherently creates and checks local person profiles, with disclosed editing behavior and manageable privacy and maintenance considerations.

Point the skill at the intended profile folder and keep backups before bulk migrations. Review saved lesson or agent-rule changes. Treat benchmark advice as examples, and separately approve account connections or publication after checking for private information. Validator success does not guarantee factual accuracy or complete formatting compliance.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (8)

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read, file_write.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The knowledge-synthesis workflow lacks a privacy warning before AI processing and public release.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The actionable advice instructs readers to consolidate three months of reading, writing, and thoughts, review the resulting knowledge base with AI, and publish selected material. It does not warn that this material may contain personal or confidential information, or require a privacy review before AI processing and publication.

Content

No source excerpt is available for this finding.

The recommended email and social-feed automation lacks a privacy warning.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The actionable recommendation instructs adopting Gmail screening and social-feed scanning in the user's agent setup, rather than merely cataloging tools. It does not warn that private messages and account content may become accessible to agents or their model providers; the earlier requirement to confirm external actions does not explain this read-access exposure.

Content

No source excerpt is available for this finding.

Automatic additions to persistent agent rules lack a system-integrity warning.

Medium
Category
Output Handling
Confidence
86% confidence
Finding

The recommendations repeatedly encourage appending AI-generated task retrospectives to AGENTS.md or skill files without specifying an approval or trust-validation boundary. If implemented literally, instructions influenced by external task content could become persistent agent rules and affect future sessions. Placement in a reference profile reduces immediate exposure and provides no authority to modify those files, but the proposed workflow still creates a conditional rule-poisoning risk.

Content

No source excerpt is available for this finding.

The writing rules require Chinese output without user language choice.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

The language-selection section explicitly requires the profile body to be written in Chinese. It provides neither user opt-in nor an exception for the user's requested language, and no locale-specific necessity is documented.

Content

No source excerpt is available for this finding.

The migration writes files before reporting changes, contrary to its promised pre-execution listing.

Medium
Category
Excessive Agency
Confidence
100% confidence
Finding

In scripts/migrate_heading_level.py, L013-L015 promise that all locations to be modified will be listed before execution. Instead, migrate_file writes immediately at L067-L068, and the caller reports only a per-file count after the write at L103-L107; no detailed pre-write listing is produced.

Content

No source excerpt is available for this finding.

The source validator enforces an 8-character description minimum despite documenting a 20-character minimum.

Low
Category
Excessive Agency
Confidence
100% confidence
Finding

In scripts/check_sources.py, L014 explicitly promises that each source description must contain at least 20 characters. However, L035 sets MIN_DESC to 8, and L096-L099 enforce that lower threshold, allowing descriptions that violate the documented acceptance criterion.

Content

No source excerpt is available for this finding.

The validator can accept duplicate required sections despite promising exactly 11 sections with no extras.

Low
Category
Excessive Agency
Confidence
99% confidence
Finding

In scripts/validate_archive.py, L014 promises an 11-section structure with neither missing nor extra sections. The implementation reduces section names to a set when checking completeness at L184-L187 and skips order validation when the count differs from 11 at L190, so a document containing all required sections plus a duplicate can pass the structure checks.

Content

No source excerpt is available for this finding.