Security audit

Optimize Context

Security checks across malware telemetry and agentic risk

Overview

The skill is not clearly malicious, but it can automatically store conversation summaries, append facts to memory, prune context, and report success while using mock messages.

Install only if you explicitly want automatic context cleanup and durable memory files. Disable scheduled cleanup and MEMORY.md updates unless needed, avoid using it on chats containing secrets or sensitive personal/business data, and verify that the command processes real session history rather than the bundled mock data.

SkillSpector

By NVIDIA

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 (14)

Intent-Code Divergence

Medium

Confidence: 93% confidence
Finding: The command is documented and presented as performing context optimization, but it operates on fabricated mock messages instead of actual session history. In a security-sensitive agent workflow, this is dangerous because it creates a false assurance that optimization, memory extraction, or cleanup occurred when no real user context was processed, which can mislead operators, break auditability, and cause downstream logic to act on incorrect state.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The skill explicitly says it 'extracts key points and facts while clearing old context' but does not warn users that prior conversation history may be discarded or summarized inaccurately. This can cause loss of important instructions, safety constraints, or user-provided data, which is a real security and reliability risk for a context-management skill whose core behavior modifies conversation state.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: The README describes a skill that automatically summarizes conversations, updates MEMORY.md, and cleans up old context, but it does not warn users about the risks of inaccurate summarization, unintended data retention, or irreversible deletion of useful context. In an agent setting, autonomous modification of memory and pruning of history can cause integrity loss, privacy issues, or operational mistakes if the system preserves the wrong facts or removes needed information.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: The skill explicitly summarizes conversation content, writes summaries and facts into persistent memory files, and clears prior context, but it provides no user-facing warning or consent checkpoint about these destructive and persistent actions. This is dangerous because users may unknowingly lose access to original conversational context or have sensitive information copied into long-lived memory artifacts, increasing the risk of data retention, corruption of memory state, and accidental disclosure.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The trigger phrase "optimize context" is broad enough that ordinary user requests about summarization or context handling could unintentionally activate this skill. Because the skill is configured to summarize and clean up conversation history, accidental activation could cause unexpected data modification or loss of conversational detail.

Vague Triggers

Low

Confidence: 81% confidence
Finding: The trigger phrase "summarize and clear" does not specify what will be cleared or under what conditions, which increases the chance of mistaken activation. In a skill that can alter stored conversation state, ambiguous activation language raises the risk of unintended cleanup behavior.

Missing User Warnings

High

Confidence: 93% confidence
Finding: The manifest explicitly states that the skill "cleans up old messages" to save tokens, but it does not present any visible warning, consent flow, or safeguard for this data-affecting behavior. Users may invoke or enable the skill without understanding that conversation history can be summarized or removed, creating risk of silent data loss and integrity issues.

Missing User Warnings

High

Confidence: 96% confidence
Finding: The schedule enables automatic daily execution and the settings enable auto-cleanup, yet the config provides no visible disclosure, approval gate, or safety control around this recurring data-modifying action. In context-management skills, unattended history cleanup is especially risky because it can repeatedly alter memory files and historical records without a contemporaneous user request.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The code appends extracted conversation facts to MEMORY.md without any consent gate, sensitivity filtering, or minimization, which creates durable storage of potentially personal or confidential user data. In an agent skill whose purpose is to process conversational context, this is especially risky because arbitrary user/assistant content can match the fact-extraction regexes and be retained beyond the original session.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The script writes conversation summaries to disk and then later clears old context, but it does not warn the user that content is being durably stored or that original context may be modified. This can cause silent persistence of sensitive prompts, preferences, tasks, or secrets in markdown files under the workspace, increasing exposure and making the agent behavior non-transparent.

Missing User Warnings

Medium

Confidence: 87% confidence
Finding: This runner invokes a context optimization flow that can persist summary data to disk and remove old summary files, but it provides no prior user warning, confirmation, or visible policy boundary around those filesystem side effects. Because the summarized content appears derived from conversation history and user preferences, the behavior can unexpectedly store potentially sensitive data and delete files in the workspace, creating privacy and integrity risks if used in an agent context.

Ssd 3

Medium

Confidence: 97% confidence
Finding: The summarization logic extracts natural-language 'facts' from user and assistant messages using broad regex patterns and prepares them for later persistent storage. Because these facts are plain-language snippets derived from arbitrary conversation content, they can capture personal details, credentials-adjacent information, or sensitive operational context, creating a durable leakage channel.

Ssd 3

Medium

Confidence: 98% confidence
Finding: The saveSummary and updateMemory paths persist both summarized conversation content and remembered facts into durable markdown files, exposing information outside the original interaction window. In this skill context, that behavior is more dangerous because the entire feature is designed to condense and preserve prior conversation, so sensitive material is systematically collected and retained rather than incidentally logged.

Autonomous Decision Making

Medium

Category: Excessive Agency
Content: - **Automatic Summarization**: Converts lengthy conversations into 10-20 key bullet points - **Fact Extraction**: Identifies and preserves important facts and information - **Memory Management**: Updates your main MEMORY.md with important facts - **Scheduled Cleanup**: Automatically runs to maintain optimal context size - **History Tracking**: Keeps track of context optimization activities - **Configurable Settings**: Uses `task_processing_config.json` for customization - **Output Optimization**: Prevents overflow and optimizes token usage
Confidence: 77% confidence
Finding: The skill advertises scheduled automatic execution that makes decisions about what information to preserve, summarize, and remove. Even without overtly malicious behavior, autonomous decision-making over conversation state is risky because incorrect heuristics can silently discard important context or store misleading summaries that affect future agent behavior.

VirusTotal

57/57 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.