Security audit

AI懂帮我巧匠

Security checks for vulnerabilities and agentic risk

Overview

This skill helps create AI profile documents, but it needs review because it encourages saving personal memory and confidential information in plaintext files while making unsupported privacy assurances.

Install only if you are comfortable with the agent creating local profile and memory documents. Do not put passwords, tokens, private keys, recovery codes, client secrets, or other reusable credentials into the generated files, and review or delete SECRET.md and MEMORY.md before sharing, syncing, or committing any output.

A.I.G

Vulnerability Patterns

Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
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 (2)

T01 · Skill Instruction Hijacking

Warning

Location: SKILL.md:95
Finding: Mandatory branded instructions override normal session output
Content: View full analysis
Remediation: View remediation

T09 · Insecure Skill Coding Practices

Error

Location: SKILL.md:578
Finding: Plaintext sensitive-information profile is encouraged under unverifiable privacy assurances
Content: View full analysis
Remediation: View remediation

SkillSpector

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 (13)

Tp2

High

Category: MCP Tool Poisoning
Confidence: 85% confidence
Finding: Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
Content: No source excerpt is available for this finding.

Ssd 3

High

Category: Not specified by scanner
Confidence: 98% confidence
Finding: The skill explicitly instructs compiling sensitive user information, long-term memory, and confidential material into dedicated files, including 'SECRET.md'. Centralizing secrets and personal profiles into reusable plaintext artifacts substantially increases the blast radius of accidental disclosure, local compromise, sync leakage, or later unintended reuse by other tools.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

High

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The skill repeatedly assures users that conversation data is not stored or uploaded, yet the workflow explicitly creates durable documents containing personal and behavioral information. This mismatch can mislead users into oversharing sensitive data under false privacy assumptions, creating consent and data-handling risk.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

High

Category: Not specified by scanner
Confidence: 98% confidence
Finding: The opening script says the skill 'won't store a single word', but later instructions require generating persistent files and long-term memory artifacts from the same conversation content. This is dangerous because users are encouraged to disclose identity, work, and preference data based on a materially inaccurate representation of storage behavior.
Content: No source excerpt is available for this finding.

Ssd 3

High

Category: Not specified by scanner
Confidence: 98% confidence
Finding: The second-stage 'AI soul archive' asks the agent to reveal stored memory, user understanding, and 'little secrets'. That creates a direct pathway for exposing accumulated sensitive information back into a user-visible artifact, where it can be copied, shared, or surfaced outside its original context.
Content: No source excerpt is available for this finding.

Ssd 3

High

Category: Not specified by scanner
Confidence: 99% confidence
Finding: The '搜魂大法' workflow operationalizes extraction into files named SOUL.md, USER.md, MEMORY.md, TOOLS.md, SECRET.md, and EMAIL_RULES.md. Formalizing that export makes sensitive aggregation routine, persistent, and easy to misuse, especially because one of the artifacts is dedicated to confidential information.
Content: No source excerpt is available for this finding.

Ssd 3

High

Category: Not specified by scanner
Confidence: 98% confidence
Finding: The templates normalize persistent storage of remembered user details, feedback history, tasks, decision records, and sensitive information for future reuse. Structured persistence of this data creates durable dossiers that can outlive the original interaction and be accessed by other processes, users, backups, or sync systems.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The description says the skill activates on phrases such as "刚注册不会用", "搜魂大法", and "看看AI的灵魂". At least some of these are broad or metaphorical enough that they could be used in casual chat without clearly intending this specific skill, increasing the risk of unintended invocation.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 96% confidence
Finding: Triggers like "刚注册不知道咋用", "AI赋能诊断", "让AI更懂我", "AI暖场", and especially "小懂" are not sufficiently specific to this skill's unique function. They overlap with common onboarding/help requests or casual references, and the manifest does not provide exclusion conditions to prevent accidental matches.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 90% confidence
Finding: The skill directs creation of multiple user-specific documents containing role, industry, preferences, memory, and secret data, but its privacy warnings are incomplete and contradicted elsewhere by overly absolute assurances. Collecting and structuring sensitive behavioral profiles without clear handling guidance increases exposure if files are later synced, shared, or accessed by others.
Content: No source excerpt is available for this finding.

Ssd 3

Medium

Category: Not specified by scanner
Confidence: 88% confidence
Finding: The onboarding script encourages disclosure of personal identity, profession, communication style, and work habits for downstream AI-facing documents. In context, this is a profiling workflow, and the danger is amplified by the inaccurate assurances that data is not stored, which undermines informed consent.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The template fixes the personal profile field to "中文" rather than asking for or inheriting the user's preferred language. This imposes a specific language/locale by default without user opt-in, which matches the stated policy-violation category.
Content: No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium

Category: Not specified by scanner
Confidence: 89% confidence
Finding: The skill is presented as a conversational document helper, but it also instructs execution of a local script to write Word documents to disk. That hidden side effect expands the trust boundary and can surprise users or operators who did not expect filesystem writes or script execution as part of the skill.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.