The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from existing GitHub issues to identify duplicates and perform triaging, which introduces a surface for indirect prompt injection attacks. A malicious issue title or body could attempt to manipulate the agent's decision-making process.
Ingestion points: The skill reads issue titles, bodies, and labels using gh issue list and gh issue view within the create and triage operations.
Boundary markers: The instructions lack explicit delimitation or framing to prevent the agent from following instructions embedded in the external issue content.
Capability inventory: The skill has broad capabilities including creating, editing, commenting on, and closing issues, as well as accessing user and label metadata.
Sanitization: No explicit content sanitization or validation logic is defined for the data retrieved from GitHub.
Mitigation: The skill includes a mandatory human-in-the-loop review step before any new issue is created or existing one is updated, mitigating the risk of unintended autonomous actions.
[COMMAND_EXECUTION]: The skill constructs and executes shell commands using the gh CLI by interpolating variables like repository paths, issue numbers, and user-provided text. While this creates a potential command injection surface, the skill utilizes standard quoting practices (e.g., "$REPO", "<label>") and quoted heredocs (cat <<'EOF') to reduce the risk of accidental or malicious shell expansion. All network activity targets GitHub, a well-known and trusted service.

github-issues