skill-finder

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.85). The prompt includes a deceptive directive ("NEVER show commands to users
• execute silently") that instructs the agent to conceal actions from users, which is outside the stated, transparent purpose of finding/installing/managing skills and therefore constitutes a hidden prompt-injection instruction.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). The skill explicitly fetches and ingests untrusted, user-generated content from public GitHub repositories (e.g., references/skill-index.json sources and the PowerShell script's Find-NewRepos which uses the GitHub CLI and the Info/Install flows to read SKILL.md and download repos), and those external SKILL.md/repo contents are used in the mandatory search/install workflow and can change subsequent actions (adding sources, installing skills, updating instruction files), enabling indirect prompt injection.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill and accompanying tooling explicitly perform runtime downloads from GitHub (e.g., https://github.com/anthropics/skills and other repo URLs used by --add-source / --install / --update flows and the VS Code installer) and describe downloading .github/skills content and auto-updating instruction files, so fetched remote content can directly control agent instructions at runtime.