Security audit

Cognitive Flexibility Release

Security checks for vulnerabilities and agentic risk

Overview

The skill is not destructive or deceptive, but it defaults to storing raw task and context data in plaintext logs and asks for broader tool access than its core reasoning purpose needs.

Review this skill before installing in environments that handle private prompts, customer data, source code, credentials, or business context. Disable monitoring where possible, avoid passing sensitive context, delete or restrict the logs directory, and consider removing unused tool permissions such as Write, Edit, sessions_send, and web_search unless the publisher documents a clear need.

A.I.G

Vulnerability Patterns

Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
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

Findings (3)

T09 · Insecure Skill Coding Practices

Warning

Location: scripts/usage_monitor.py:58
Finding: Default Monitoring Persists Complete Task and Context Data in Plaintext
Content: View full analysis
= 0.7, "needs_improvement": result.get("assessment", {}).get("needs_improvement", False), "mode_switched": result.get("mode_switched", False), "previous_mode": result.get("previous_mode", None), "context": context or {} } # Write to log file with open(self.log_file, 'a', encoding='utf-8') as f: f.write(json.dumps(log_entry, ensure_ascii=False) + '\n') ``` ### Technical Analysis `CognitiveController` enables monitoring by default and sends the complete task and context to `UsageMonitor.log_usage()`. The monitor serializes both values directly into `logs/usage-log.jsonl`. No data minimization, secret filtering, explicit user consent, retention policy, encryption, or restrictive file-permission handling is applied. The `task_hash` field does not protect th ...[truncated 1266 chars]
Remediation: View remediation

T01 · Skill Instruction Hijacking

Warning

Location: scripts/pattern_matcher.py:61
Finding: Untrusted Memory Content Can Be Propagated Verbatim into the Final Answer
Content: View full analysis
List[Dict]: """Retrieve similar tasks""" similar_tasks = [] # Search through memory_search if tools and hasattr(tools, 'memory_search'): memories = await tools.memory_search(query=task, maxResults=10) similar_tasks = memories.get('results', []) # Filter highly similar tasks filtered = [] for task_item in similar_tasks: similarity = self._calculate_similarity(task, task_item) if similarity >= self.similarity_threshold: task_item['similarity'] = similarity filtered.append(task_item) # Sort by similarity filtered.sort(key=lambda x: x.get('similarity', 0), reverse=True) return filtered[:5] ``` ```python # scripts/pattern_matcher.py:84-99 def _extract_patterns(self, similar_tasks: List[Dict]) -> List[Dict]: """Extract patterns from similar tasks""" patterns = [] for task in similar_tasks: # Extract task pattern pattern = { "task_type": task.get("task_type", "unknown"), "solution_approach": task.get("solution_approach", ""), "success_factors": task.get("success_factors", []), "pitfalls": task.get("pitfalls", []), "similarity": task.get("similarity", 0) } patterns.append(pattern) return patterns ``` ```python # scripts/pattern_matcher.py:153-164 def apply_pattern(self, pattern: Dict, current_task: str) -> Dict: """Apply a pattern to the current task""" return { "original_task": current_task, "applied_pattern": pattern, "adapted_solution": self ...[truncated 3157 chars]
Remediation: View remediation

T05 · Unauthorized Access and Privilege Escalation

Note

Location: SKILL.md:20
Finding: Declared Agent Tool Permissions Exceed Implemented Functional Requirements
Content: View full analysis
Remediation: View remediation

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
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

Findings (44)

Tp4

High

Category: MCP Tool Poisoning
Confidence: 99% confidence
Finding: The file advertises complete usage logs, statistics, monitoring, and reporting, but it does not warn about what task details may be persisted, for how long, or whether sensitive prompts/context are included. Undisclosed persistent logging can expose confidential user data on disk and create secondary leakage through reports or retained analytics.
Content: No source excerpt is available for this finding.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 93% confidence
Finding: The file advertises complete usage logs, statistics, monitoring, and reporting, but it does not warn about what task details may be persisted, for how long, or whether sensitive prompts/context are included. Undisclosed persistent logging can expose confidential user data on disk and create secondary leakage through reports or retained analytics.
Content: No source excerpt is available for this finding.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 95% confidence
Finding: The file advertises complete usage logs, statistics, monitoring, and reporting, but it does not warn about what task details may be persisted, for how long, or whether sensitive prompts/context are included. Undisclosed persistent logging can expose confidential user data on disk and create secondary leakage through reports or retained analytics.
Content: No source excerpt is available for this finding.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 95% confidence
Finding: The file advertises complete usage logs, statistics, monitoring, and reporting, but it does not warn about what task details may be persisted, for how long, or whether sensitive prompts/context are included. Undisclosed persistent logging can expose confidential user data on disk and create secondary leakage through reports or retained analytics.
Content: No source excerpt is available for this finding.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 98% confidence
Finding: The file advertises complete usage logs, statistics, monitoring, and reporting, but it does not warn about what task details may be persisted, for how long, or whether sensitive prompts/context are included. Undisclosed persistent logging can expose confidential user data on disk and create secondary leakage through reports or retained analytics.
Content: No source excerpt is available for this finding.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 97% confidence
Finding: The file advertises complete usage logs, statistics, monitoring, and reporting, but it does not warn about what task details may be persisted, for how long, or whether sensitive prompts/context are included. Undisclosed persistent logging can expose confidential user data on disk and create secondary leakage through reports or retained analytics.
Content: No source excerpt is available for this finding.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 91% confidence
Finding: The file advertises complete usage logs, statistics, monitoring, and reporting, but it does not warn about what task details may be persisted, for how long, or whether sensitive prompts/context are included. Undisclosed persistent logging can expose confidential user data on disk and create secondary leakage through reports or retained analytics.
Content: No source excerpt is available for this finding.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 98% confidence
Finding: The file advertises complete usage logs, statistics, monitoring, and reporting, but it does not warn about what task details may be persisted, for how long, or whether sensitive prompts/context are included. Undisclosed persistent logging can expose confidential user data on disk and create secondary leakage through reports or retained analytics.
Content: No source excerpt is available for this finding.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 99% confidence
Finding: The file advertises complete usage logs, statistics, monitoring, and reporting, but it does not warn about what task details may be persisted, for how long, or whether sensitive prompts/context are included. Undisclosed persistent logging can expose confidential user data on disk and create secondary leakage through reports or retained analytics.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: The document explicitly describes ongoing usage monitoring and daily feedback collection channels, but it provides no notice about what data is collected, how long it is retained, who can access it, or whether users can opt out. In a skill intended for broad community release, this creates a real privacy and compliance risk because telemetry and feedback artifacts may include user prompts, model outputs, identifiers, or operational metadata.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The token-based login instructions tell the user to create, copy, and place a CLI token into a command or environment variable, but they do not warn that the token is a secret or that it can be exposed through shell history, screenshots, terminal recording, shared transcripts, or persisted environment state. If mishandled, the token could allow unauthorized publication or account actions in ClawHub under the user's identity.
Content: No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The guide introduces persistent monitoring, usage tracking, and effectiveness evaluation features that go beyond the stated purpose of a cognitive reasoning/flexibility skill. Even without executable code shown here, documenting collection of task, mode, result, context, timestamps, and reports expands the skill into telemetry and data collection, increasing privacy and data-governance risk.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The monitoring guide describes logging and storage behavior but omits any warning that user task data may be collected and persisted. This lack of transparency increases the chance that operators unknowingly store sensitive information, creating privacy, compliance, and insider-access risks.
Content: No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium

Category: Not specified by scanner
Confidence: 92% confidence
Finding: Detailed usage logging and report/export capability is not necessary for core cognitive-flexibility behavior and creates a side-channel for collecting operational and potentially sensitive task data. Because the documented API logs task and context, this can capture user prompts, business data, or internal workflows without clear limitation.
Content: No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: Task-level log export materially increases risk because it creates an easy data extraction path for timestamps, modes, and task contents into a portable JSON file. Exported files are easier to copy, mishandle, or ingest elsewhere, which magnifies exposure of sensitive prompts and contextual information beyond the original runtime environment.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The export example explicitly shows iterating over logs containing timestamps, mode, and task fields, yet provides no caution about sensitive data in the resulting JSON file. This is dangerous because exported artifacts commonly escape normal access controls and can be shared, backed up, or exposed accidentally.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: 该发布指南全文以中文编写，未说明这是面向特定中文受众的区域化文档，也未提供其他语言选项或用户选择机制。根据规则，强制特定语言而无用户选择或明确合理化说明，属于自然语言层面的语言/区域策略违规。
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The skill description is written as a directive in Chinese and the README provides no indication that other languages are supported or that the language choice is optional. Under the language/locale policy, forcing a specific language without user opt-in is a natural-language policy violation.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 92% confidence
Finding: The activation guidance is broad enough to encourage use on many normal tasks, which increases unnecessary exposure to powerful tools like memory_search, sessions_send, web_search, Write, and Edit. Overbroad invocation criteria can lead operators or automated planners to select this skill in contexts where its data access and modification capabilities are not needed.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The skill explicitly advertises usage tracking, monitoring, stats, and report generation, yet gives no warning about potential data collection, retention, or sensitivity of logged task context. In a reasoning skill likely to process proprietary or personal inputs, silent telemetry materially raises privacy and data-handling risk.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 90% confidence
Finding: This manifest mixes English metadata with Chinese-only descriptive fields and labels, including the top-level heading and description text. For a general-purpose skill with no documented regional or locale constraint, forcing a specific language in user-facing natural language can violate language/locale policy when no user opt-in or choice is offered.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 92% confidence
Finding: The file explicitly describes automatic collection of usage metrics and provides commands to export logs, but it does not mention consent, minimization, retention, access controls, or any user-facing privacy notice. In a skill context, usage logs can contain prompts, identifiers, or behavioral metadata, so undocumented telemetry can create privacy and compliance risk even if the feature is intended for benign analytics.
Content: No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The manifest describes a skill for cognitive flexibility, mode switching, metacognitive monitoring, and complex reasoning support. This file instead implements persistent user-feedback collection, logging, statistics generation, and export functionality, which is an operational telemetry/maintenance capability rather than a direct requirement of the stated reasoning purpose.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Medium

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The module documentation says it 'automatically collects user feedback' and 'collects from various channels' for the skill. In practice, the implementation only provides add_feedback(feedback) to append caller-supplied dictionaries to local files; there is no code for channel integration, polling, or automatic collection from GitHub, Discord, email, or other sources.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: The code writes full feedback objects, including user identifiers and free-form descriptions, directly to a local JSONL log without minimization, consent handling, retention controls, or access restrictions. In this skill context, the feedback system is ancillary to the core purpose, so collecting and persisting potentially sensitive user content expands privacy exposure without clear necessity.
Content: No source excerpt is available for this finding.