The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests and processes content from external GitHub and GitLab issues.
Ingestion points: External data is ingested through gh issue view and glab issue view commands as described in the issue tracker reference files.
Boundary markers: A structured 'Changeset' format is employed to stage mutations, acting as a logical boundary between the analysis and execution phases.
Capability inventory: The skill has capabilities to perform file writes and execute repository management commands (gh, glab, git).
Sanitization: While no explicit content sanitization is mentioned, the skill mandates a human-in-the-loop approval gate ('Halt turn immediately for human approval') before any staged changes are written to disk, which significantly mitigates the risk of automated exploitation.

zenforge-init