The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill processes untrusted data from GitLab resources such as issue titles, merge request descriptions, and code blobs. An attacker with access to a GitLab repository could inject malicious instructions into these fields to influence the agent's behavior when it searches those resources.
Ingestion points: Results from glab api "search?scope=..." calls.
Boundary markers: Absent. The skill does not implement delimiters or 'ignore' instructions for the data it retrieves.
Capability inventory: The skill allows execution of glab commands and bash subshells, providing a platform for further actions if an injection is successful.
Sanitization: Absent. Data returned from the GitLab API is processed directly by jq and presented to the agent context.
[Data Exposure] (LOW): Workflow 4 explicitly demonstrates searching for sensitive strings like 'password', 'secret', and 'api_key'. While this is a common use case for security auditing, it actively brings sensitive credentials into the agent's working memory.
[Command Execution] (SAFE): The skill uses glab api and bash for its primary functions. It uses jq for URI encoding (jq -Rr @uri) to safely construct URLs from user input, which is a defensive best practice for preventing command or path injection within the API calls.

gitlab-search