Agent-Skills-for-Context-Engineering

• Outcome: Pass
• Latest audit: Jul 10, 2026 · 2:36 AM UTC
• Version: 1.0.0

Overview

This is a coherent context-compression skill whose main risk is preserving sensitive conversation details in summaries, not hidden or malicious behavior.

Reasonable to install for context-compression workflows. Treat generated summaries and evaluation inputs as potentially sensitive because they can preserve user goals, file paths, errors, decisions, and excerpts from prior conversation. If adapting the LLM-judge example, redact sensitive content and get explicit user control before sending compressed context to any external provider.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Memory Poisoning: Persistent Context Injection, Context Window Stuffing, Memory Manipulation
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Vague Triggers

Medium

• Confidence: 91% confidence
• Finding: The manifest description mixes specific phrases like "compress context" with broader conditions such as "reduce token usage" and especially "mentions context compression" or "long-running agent sessions exceeding context limits." These conditions are not tightly scoped to explicit invocation phrases and could overlap with general discussion of tokens or session length, making activation boundaries unclear.

Vague Triggers

Low

• Confidence: 84% confidence
• Finding: The "When to Activate" section lists several scenarios such as "Designing conversation summarization strategies" and "Building evaluation frameworks for compression quality," but it does not clarify when the skill should not activate. Without exclusions or a narrower scope, these broad topical descriptions may trigger the skill during general discussion rather than intentional use.

Context Window Stuffing

Medium

• Category: Memory Poisoning

Content

# Context Compression Strategies

When agent sessions generate millions of tokens of conversation history, compression becomes mandatory. The naive approach is aggressive compression to minimize tokens per request. The correct optimization target is tokens per task: total tokens consumed to complete a task, including re-fetching costs when compression loses critical information.

## When to Activate

• Confidence: 80% confidence
• Finding: generate millions of tokens

Context Window Stuffing

Medium

• Category: Memory Poisoning

Content

## When to Activate

Activate this skill when:
- Agent sessions exceed context window limits
- Codebases exceed context windows (5M+ token systems)
- Designing conversation summarization strategies
- Debugging cases where agents "forget" what files they modified

• Confidence: 80% confidence
• Finding: exceed context window

Context Window Stuffing

Medium

• Category: Memory Poisoning

Content

Activate this skill when:
- Agent sessions exceed context window limits
- Codebases exceed context windows (5M+ token systems)
- Designing conversation summarization strategies
- Debugging cases where agents "forget" what files they modified
- Building evaluation frameworks for compression quality

• Confidence: 80% confidence
• Finding: exceed context window

VirusTotal

66/66 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.