The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous templates for executing shell commands using the glab CLI. It identifies shell injection risks associated with handling content with special characters like backticks and provides specific mitigations, such as using single-quoted heredocs and temporary files. Additionally, the skill documentation uses dynamic context injection (e.g., !glab mr create --help) to include CLI help text at load time, which is considered benign usage.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to its interaction with untrusted external data sources.
Ingestion points: Untrusted data enters the agent's context through commands like glab api and glab issue view, as described in SKILL.md and scripts/epic-notes.sh.
Boundary markers: The skill recommends using single-quoted heredoc delimiters (<< 'EOF') to isolate content when writing data, serving as a defense against accidental shell expansion.
Capability inventory: The skill can execute shell commands (glab), write files (via mktemp and output redirection), and perform network operations through the GitLab API.
Sanitization: Instructions emphasize the use of GraphQL variables and safe shell syntax to ensure that untrusted content is handled as literal data rather than executable code.

glab