The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub issues and comments, creating a surface for indirect prompt injection attacks where an attacker could embed instructions in an issue to manipulate the triaging agent.
Ingestion points: The skill uses gh issue view in SKILL.md to fetch external content including issue titles, bodies, and comments from the kube-hetzner/terraform-hcloud-kube-hetzner repository.
Boundary markers: No specific delimiters or instructions to ignore embedded commands are present in the workflow or templates when the agent processes the fetched content.
Capability inventory: The skill utilizes the GitHub CLI (gh) to perform actions such as editing labels (gh issue edit), closing issues (gh issue close), and transferring issues to discussions (gh issue transfer).
Sanitization: While the instructions suggest users provide sanitized configuration files, there are no explicit sanitization, filtering, or validation steps applied to the retrieved issue text itself before analysis.

triage-issue