The Agent Skills Directory

[SAFE]: The skill is authored by PrefectHQ and focuses on managing its own public repository (PrefectHQ/prefect). All commands and target resources are consistent with the vendor's legitimate operations.
[SAFE]: The skill enforces a 'read-only' default mode, requiring explicit user approval before any edits or labels are applied to the repository. This human-in-the-loop constraint effectively prevents autonomous or accidental destructive actions.
[SAFE]: Command usage is limited to the official GitHub CLI (gh) for listing issues and labels. These are non-privileged operations that do not involve downloading or executing untrusted code.
[PROMPT_INJECTION]: The skill ingests untrusted data from GitHub issue descriptions and comments (Category 8: Indirect Prompt Injection surface).
Ingestion points: Untrusted data enters the agent context via gh issue list in SKILL.md.
Boundary markers: No explicit delimiter strings are defined, but the multi-pass review process serves as a functional boundary.
Capability inventory: Capabilities are restricted to issue classification and labeling; no access to sensitive system files or arbitrary network operations is present.
Sanitization: The skill uses predefined templates for comments, ensuring the agent's output is structured and safe regardless of input content.

backlog-management