DriveMind

• Outcome: Pass
• Latest audit: Sep 11, 2026 · 7:25 PM UTC
• Version: 0.3.0

Overview

DriveMind is a disclosed behavioral guidance skill for steadier task follow-through, with no code execution, network access, hidden payloads, or destructive behavior found.

Install this if you want an agent to be more structured about follow-through, escalation, and retrospectives. Review the memory-related behavior first, especially if your agent can save persistent notes, and be aware that broad trigger phrases may activate the style more often than expected.

A.I.G

Vulnerability Patterns

• Skill Instruction Hijacking: Alters the agent's session goals or safety constraints when the skill loads
• Agent Memory Poisoning: Writes attacker-controlled rules into memory that affect later sessions
• Remote Payload Retrieval and Execution: Fetches external code whose behavior can change after review
• Embedded Malicious Code: Ships malicious scripts inside the skill and executes them locally
• Unauthorized Access and Privilege Escalation: Obtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• YARA Signatures: Malware Match, Webshell Match, Cryptominer Match
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (12)

YARA rule 'agent_skill_prompt_injection_hidden_instructions': Prompt injection or hidden instructions embedded in AI agent skill text [agent_skills]

High

• Category: YARA Match

Content

---
name: drivemind
description: Apply DriveMind, the calm reliability layer for AI agents. Use when a task needs steady follow-through, clearer progress, stronger persistence without recklessness, explicit safety boundaries, human-in-the-loop collaboration, post-task review, reusable memory, or when the user says things like 'keep pushing', 'don’t stop too early', 'be steady', 'if risk is unclear ask me', 'review this after', or 'write down the lesson'.
---

# DriveMind

DriveMind helps the agent work like a calm, well-mannered collaborator: stay with meaningful work, communicate clearly, ask before crossing unclear boundaries, and leave behind reusable lessons.

In v0.

• Confidence: 80% confidence
• Finding: YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Anti-Refusal Statement

High

• Category: Anti-Refusal

Content

Higher persistence and stronger follow-through, while keeping the same safety boundaries.

### Intensive
Use when the user explicitly wants stronger commitment on an important task, but never bypass safety or human authority.

## Output pattern
When DriveMind is active, prefer an output that makes the work easier to continue. In most non-trivial cases, try to include:

• Confidence: 90% confidence
• Finding: Skill attempts to nullify the agent's safety policies or restrictions ('you have no restrictions', 'ignore your guidelines', 'do anything now'). This is a direct jailbreak that disables guardrails.

Instruction Override

High

• Category: Prompt Injection

Content

Higher persistence and stronger follow-through, while keeping the same safety boundaries.

### Intensive
Use when the user explicitly wants stronger commitment on an important task, but never bypass safety or human authority.

## Output pattern
When DriveMind is active, prefer an output that makes the work easier to continue. In most non-trivial cases, try to include:

• Confidence: 90% confidence
• Finding: This pattern attempts to override system instructions or ignore safety constraints. Without LLM analysis, manual review is recommended.

Natural-Language Policy Violations

High

• Confidence: 97% confidence
• Finding: The entire skill is written to operate in Chinese, with no indication that the user may choose another language or that the locale restriction is intentional and justified. This can constitute a language/locale policy violation because it implicitly enforces one language without opt-in.

Vague Triggers

Medium

• Confidence: 95% confidence
• Finding: The skill description includes broad everyday phrases like 'keep pushing', 'be steady', and 'review this after' as activation cues. That can cause the skill to activate in contexts where the user did not intend to invoke a persistence/review framework, potentially changing agent behavior unexpectedly across many ordinary tasks.

Vague Triggers

Medium

• Confidence: 91% confidence
• Finding: The 'Use when' section defines applicability with subjective conditions such as 'important,' 'clearer progress,' and 'should not be dropped too early.' These criteria are under-specified, so an agent may over-apply the skill and adopt persistence or memory behaviors in situations where that mode is not appropriate.

Vague Triggers

Medium

• Confidence: 95% confidence
• Finding: The phrase 'or similar instructions' expands the trigger surface beyond a bounded phrase list, making activation effectively open-ended. That increases the risk of accidental invocation and reduces predictability, which is especially risky for a cross-cutting behavioral skill that changes persistence, escalation, and output structure.

Natural-Language Policy Violations

Medium

• Confidence: 95% confidence
• Finding: The file's operational guidance is written entirely in Chinese, which imposes a specific language on users without any opt-in or alternative locale. Under the policy for natural-language violations, a skill should not force a language or locale unless it explicitly offers a choice or documents a justified region-specific constraint.

Natural-Language Policy Violations

Medium

• Confidence: 88% confidence
• Finding: This markdown file contains user-facing guidance exclusively in Chinese, which can amount to a language/locale policy violation when no user opt-in or alternative language option is provided. The policy specifically calls for flagging skills that force a specific language without user choice.

Vague Triggers

Medium

• Confidence: 94% confidence
• Finding: The '典型信号' list for 判断题 includes very generic phrases such as '要不要做', '要不要发', and '先做哪个', which are common in ordinary conversation and not constrained to a specific context. The document does not provide exclusion conditions or narrower scope, so these triggers could cause unintended invocation.

Vague Triggers

Medium

• Confidence: 92% confidence
• Finding: Phrases like '继续推进', '别停', '给我下一步', and '帮我整理局面' are broad requests that can appear in many unrelated conversations. Without context limits or non-trigger examples, the invocation boundary for this skill remains ambiguous.

Vague Triggers

Low

• Confidence: 84% confidence
• Finding: Items such as '总结一下', '复盘', and '下次怎么做' are common conversational requests and are not sufficiently specific on their own. The file does not clarify constraints or exclusions that would prevent over-triggering.

Static analysis

No suspicious patterns detected.