Back to skill

Security audit

Apollo Dream

Security checks for vulnerabilities and agentic risk

Overview

This memory-management skill is not plainly malicious, but it can automatically save conversation-derived decisions, preferences, and snapshots to local files with loose scoping and limited user control.

Review before installing. Use it only if you want an agent to retain selected conversation history, decisions, and preferences in local memory files. Prefer an isolated workspace, inspect the generated .dream and .workflow/snapshots data, avoid giving it sensitive files as snapshot inputs, and require explicit approval before persistent memory writes or automatic trigger handling.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (21)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

声明描述的是一种广义的记忆整理与遗忘机制,重点应是筛选重要与不重要的信息;但代码并未实现记忆压缩、重要性评估或遗忘逻辑。相反,它专门检测特定语言模式中的决策语句,并把这些决策记录到 /root/.openclaw/workspace/.dream/decisions.json。其主要目的和实际能力都更接近“决策跟踪/记录”而不是“梦境式记忆整理”。因此描述与行为存在明显不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是“整理记忆、保留重要内容、忘掉不重要内容”的记忆管理或内容筛选功能,但实际代码没有任何记忆整理、筛选、摘要、删除或重要性判断逻辑。它仅对输入文本或文件进行 token、字符数、行数的估算,并从本地文件读取内容进行统计。这与声明的主要用途明显不符,属于实质性功能不匹配。

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The v1整理流程 explicitly includes '冲突检测 + 去重' (Step 3), checking links with old memories, and automatic task/topic state decisions via scripts (Step 5). But the subsequent 'v1 不做的事(v2+)' section says those same capabilities are not part of v1, creating a direct contradiction in the skill's own implementation guidance.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
89% confidence
Finding

The skill includes a direct shell deletion command against a path in /tmp as part of its routine flow. Even though the specific target is narrow, embedding raw destructive commands in a natural-language skill increases the chance of unsafe tool use, path confusion, or unintended deletion if the command is generalized, templated, or executed without validation.

Content

Scanner excerpt · SKILL.md (reported line 245)May include surrounding context.

→ 保留:结论 + 任务状态快照 + 索引

Step 7: 清理trigger → rm /tmp/.dream-trigger → 通知主AI整理完成

text

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding

The skill describes capabilities that read and write local files but does not declare any explicit tool scope or permission boundary. That makes its effective authority unclear, which can lead to unintended file access or modification when the skill is invoked in contexts broader than the user expects.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill is configured to activate on broad conditions like long conversations, low token budget, or vague needs to organize context. Such ambiguous triggers increase the chance of unsolicited execution, which is risky here because the skill can persist conversation-derived data and modify local state.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Generic trigger phrases like '做梦', '记忆整理', or '上下文管理' are common in ordinary conversation and may invoke the skill unintentionally. Because the skill performs persistence and cleanup actions, accidental invocation can cause unauthorized data retention or state changes.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill instructs the AI to write extracted information to files and delete or discard details, but it does not provide a clear user-facing warning that local data will be modified. Hidden persistence and cleanup behavior can violate user expectations and lead to unauthorized storage or loss of conversation-derived information.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill explicitly directs the AI to retain user-related conversation content, key decisions, and preferences in persistent memory files across sessions. Persisting user-derived data without clear minimization, consent, retention limits, or access controls creates a privacy and data-handling risk.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

The '待实现的(未来工作)' section states that automatic memory association discovery is not yet implemented. However, later workflow steps direct the system to check for links to old memories and perform conflict detection/deduplication, which presents those capabilities as active behavior rather than future work.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The post-task flow tells the AI to extract unresolved issues and memory links from conversation content and save them into long-term files. This increases the sensitivity of stored artifacts because it turns transient interaction data into structured, reusable profiles or histories.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The v1 workflow requires scanning conversation history, extracting preferences and entities, and writing the results into long-term memory artifacts. In context, this is more dangerous because it combines broad automatic triggering with structured persistence of potentially sensitive user information, enabling silent profile building across sessions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code file contains natural-language descriptions and user-facing output entirely in Chinese, including the module description and operational labels. Under the policy rule, forcing a specific language without user opt-in is a locale/language policy violation when no choice or justification is provided.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The script persistently stores extracted decision content to a file under /root/.openclaw/workspace/.dream/decisions.json without any consent prompt, disclosure, retention control, or access restriction logic. Because the tracked text may contain sensitive user decisions, preferences, or operational context, silent persistence creates a privacy and data exposure risk, especially in a memory-oriented skill whose purpose is to retain selected conversation content.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code file contains natural-language descriptions and command-line output exclusively in Chinese, including the module docstring and usage/help text. Under the policy, forcing a specific language without offering a choice or documenting a justified locale constraint is a language-policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code file contains natural-language strings that force a specific language/locale for users, including the module docstring and CLI usage/help text. Under the policy, language-specific behavior is only acceptable when users are given a choice or the locale restriction is clearly documented and justified.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The read_snapshot function accepts any caller-supplied path and opens it directly without restricting access to the snapshots directory. In the CLI, this allows a user or invoking agent to read arbitrary files on disk that are valid JSON, which can expose secrets or internal state outside the intended snapshot mechanism.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The create CLI command reads any file path provided by the caller and stores its contents as a snapshot. In an agent skill context, this can be abused to exfiltrate local sensitive files into the workflow snapshot store, making secrets easier to persist, inspect, or forward later.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This Python file contains natural-language docstrings and CLI messages entirely in Chinese, including the script description and usage/help text. Under the policy rule, forcing a specific language without user opt-in is a locale/language policy violation unless the constraint is explicitly justified, which is not present here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The description and operational instructions are presented entirely in Chinese, and the file does not indicate that language choice is optional or that the skill is intended only for a Chinese-language environment. This can amount to an implicit locale constraint without opt-in under the policy's language-choice requirement.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The file's descriptive text and all user-facing CLI output are written in Chinese, which imposes a specific language on users without any opt-in or alternative. The policy allows locale constraints only when documented and justified or when users are given a choice, neither of which is present here.

Content

No source excerpt is available for this finding.