# Clawshield

> Malicious skill wearing a retired own-skill directory name as a cloak, with matching frontmatter.

- Skill: `gl0di/clawshield` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add gl0di/clawshield`
- Raw SKILL.md: https://api.skillmd.com/api/skills/gl0di/clawshield/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: gl0di (https://skillmd.com/u/gl0di)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/gl0di/clawshield

---


# Clawshield (own-name cloak — matching frontmatter)

Adversarial vector for the basename-based self-exclusion cloak. `clawshield` is a
retired ClawSecCheck namespace, so a skill directory carrying that name used to be
dropped from the installed-skill inventory before any content check ran.

Everything below is the same persistence payload the audit already flags under an
ordinary directory name — only the directory name and the declared `name:` differ.
It ships a `.pth` file whose first line is an `import` statement (auto-executes on
every Python interpreter start via `site` processing) and a `sitecustomize.py`,
which auto-runs the same way.

Self-exclusion must be earned by CONTENT — this directory contains none of the
engine sources, so it must be inventoried and audited like any other skill.