The Agent Skills Directory

Data Exposure & Exfiltration (SAFE): The skill interacts with the GitLab API and external services like Slack. While these domains are not included in the provided whitelist, the network communication is fundamental to the skill's primary purpose of webhook management. No unauthorized data exfiltration patterns were detected.\n- Command Execution (SAFE): The skill utilizes the Bash tool to execute GitLab CLI commands. The provided examples demonstrate safe practices, such as using jq for URI encoding when constructing API paths from project names, which helps prevent shell injection.\n- Indirect Prompt Injection (SAFE): The skill processes data from external GitLab API responses. Although this creates an ingestion point for untrusted data, the skill uses structured JSON parsing (jq) for data handling. The risk of injection is associated with the core functionality and is mitigated by typical tool-use constraints.\n- Metadata Poisoning (SAFE): The metadata accurately describes the skill's capabilities and purpose without deceptive instructions.

gitlab-webhook