The Agent Skills Directory

Prompt Injection (SAFE): No evidence of role-play, DAN-style instructions, or system prompt extraction was found. The instructional language is standard for a developer tool.
Data Exposure & Exfiltration (SAFE): Example files in the references/ directory use appropriate placeholders (e.g., your-api-key, change-me-in-production). No commands that exfiltrate sensitive files (like ~/.aws/credentials) to external domains were detected.
Unverifiable Dependencies & Remote Code Execution (SAFE): The skill utilizes standard tools like jq, git, and gh. Remote code execution patterns are limited to pulling a Docker image from a trusted registry (ghcr.io/typo3-documentation/render-guides:latest) for documentation rendering.
Obfuscation (SAFE): No multi-layer Base64, zero-width characters, or homoglyph attacks were found. Integrity hashes in lockfiles are standard and non-malicious.
Privilege Escalation (SAFE): No usage of sudo, chmod 777, or dangerous permission changes was detected.
Indirect Prompt Injection (LOW): The skill ingests untrusted project metadata (e.g., package names, script descriptions) to generate documentation. While this is an inherent attack surface for documentation generators, the skill uses standard parsing techniques and does not have high-privilege write capabilities outside of generating the AGENTS.md file itself.

agents