The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The installation scripts fetch repository configurations and packages from official GitLab domains.
Evidence: scripts/install.sh and scripts/install.ps1 target https://repo.gitlab.com to configure the system package manager.
[REMOTE_CODE_EXECUTION]: The skill utilizes a remote execution pattern for automated repository setup on Linux distributions.
Evidence: scripts/install.sh executes a script directly from repo.gitlab.com using the curl | sudo bash pattern.
[COMMAND_EXECUTION]: The skill performs shell command execution to interact with the glab CLI and manages packages via sudo during the installation phase.
Evidence: scripts/glab-auth-status.sh and scripts/glab-preflight.sh execute glab commands; scripts/install.sh uses sudo apt install and sudo dnf install.
[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by reading untrusted, user-generated content from GitLab into the agent's context.
Ingestion points: The skill reads merge request descriptions, issue comments, and CI logs via glab mr view, glab issue list, and glab ci trace as documented in references/command-cookbook.md.
Boundary markers: Absent. No specific delimiters are used to separate GitLab data from agent instructions.
Capability inventory: The skill can perform sensitive operations including creating/closing issues, merging code, and uploading release assets.
Sanitization: No specific sanitization or instruction-filtering is applied to the data retrieved from GitLab.
[SAFE]: The skill implements proactive measures to protect credentials by sanitizing terminal output.
Evidence: scripts/glab-auth-status.sh uses sed -E '/[Tt]oken:/d' to ensure that GitLab Personal Access Tokens (PATs) are not included in the diagnostic logs or displayed to the user.

glab