ClawHub

Codebase Survey

Security checks across malware telemetry and agentic risk

Overview

This markdown-only skill guides a read-only codebase survey and includes explicit cautions to avoid secrets.

Install this only if you want an agent to inspect and summarize the intended repository. Keep it scoped to the project you mean to survey, and do not approve reading local/private context such as CLAUDE.local.md or secret files unless you are comfortable with that information being used.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Vague Triggers

Medium
Confidence
89% confidence
Finding
The skill declares multiple broad natural-language trigger phrases such as 'deep dive', 'explore this codebase', and 'walk me through this repo', which can match common user requests beyond a narrowly scoped invocation boundary. This can cause unintended activation of the skill, leading the agent to perform expansive repository reconnaissance or follow a heavyweight workflow when the user may have intended a smaller, more specific action.

Vague Triggers

Medium
Confidence
84% confidence
Finding
The 'When to Use' section is permissive and overlaps with adjacent activities like onboarding, planning, refreshes, and handoff context, without defining strict boundaries for when the skill should not auto-activate. In practice, this ambiguity increases the chance that normal conversational requests are interpreted as authorization to inspect large portions of a codebase, which can over-collect context and expose more project information than necessary for the user's actual need.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal