The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the custom_headers feature, which supports valueFromCommand. This allows the configuration of an external executable that runs once per glab process to provide header values, effectively serving as a command execution hook within the tool's lifecycle.
[DYNAMIC_EXECUTION]: The valueFromCommand source allows for the execution of arbitrary scripts or binaries. The documentation notes that if shell expansion is required, a shell must be invoked explicitly, which increases the risk of command injection if user-supplied data is incorporated into the configuration.
[EXTERNAL_DOWNLOADS]: The skill manages configuration for duo_cli_auto_download and orbit_cli_auto_download. These settings enable the glab CLI to automatically fetch and execute GitLab-provided binaries without interactive user confirmation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources including environment variables (e.g., GITLAB_TOKEN, GLAB_CONFIG_DIR) and local repository configuration files (.git/glab-cli/config.yml). An attacker who can influence these files or the environment could potentially manipulate the agent's behavior when it manages or reads the configuration.
Ingestion points: Environment variables and local .git configuration files (documented in SKILL.md).
Boundary markers: None; configuration keys are processed as raw strings.
Capability inventory: The skill provides full access to glab config set, get, and edit, which can modify global and local tool behavior.
Sanitization: glab performs basic validation against a canonical schema, but does not sanitize the contents of command-based header sources or token values.

glab-config