ClawHub
Back to skill

Security audit

context-budget

Security checks for vulnerabilities and agentic risk

Overview

This skill is a focused context-budgeting helper with disclosed local file reads, optional loopback tokenization, and safety guidance for user-directed instruction edits.

Install only if you want help auditing or reducing always-loaded instruction files. Treat any edits to agent instructions as high-impact changes, use the default redacted reports unless you intentionally need content shown, and only point the tokenizer option at a trusted local loopback service that does not retain requests.

A.I.G

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

SkillSpector

Vulnerability Patterns
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (3)

Lp3

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding
Without declared permissions the skill's intent is opaque and cannot be validated.

Persistent Context Injection

Medium
Category
Memory Poisoning
Content
reduce it without weakening authority, safety, or meaning.

Use this skill when runtime logs report bootstrap truncation, per-turn context
cost is high, policies appear in several files, or a persistent rule is being
added.

## The two units differ
Confidence
80% confidence
Finding
Skill injects content designed to persist in agent memory or context across interactions. Persistent injection can alter agent behavior long after the initial interaction.

Unbounded Output

Medium
Category
Output Handling
Content
3. semantic and precedence review finds no weakened or unavailable rule;
4. all six assurance regressions pass;
5. before/after characters and tokens are recorded, with estimates labelled;
6. one fresh session runs without truncation or behaviour regressions;
7. temporary backups are securely removed after the clean-session gate.
Confidence
60% confidence
Finding
Output size or generation rate is not bounded. Unbounded output enables denial-of-service through resource exhaustion, log flooding, or context-window stuffing.

Static analysis

No suspicious patterns detected.