The Agent Skills Directory

[SAFE]: The skill provides a structured framework for long-horizon autonomous development tasks. It uses an internal state file (.forge/loop-state.json) to manage progress across multiple sessions and issues.
[COMMAND_EXECUTION]: The instructions utilize standard, well-known development tools including git and the GitHub CLI (gh) for common operations such as managing worktrees, updating labels, and polling CI status. These commands are constrained to the context of the current repository.
[INDIRECT_PROMPT_INJECTION]: The orchestrator reads external project files such as forge.toml and CLAUDE.md to configure its behavior. It also processes issue content retrieved via other sub-agents. These are standard inputs for a development-focused agent and are handled within a defined state machine.
[REMOTE_CODE_EXECUTION]: While the skill orchestrates code changes and CI triggers, it does so through controlled sub-agents (/autodev, /review-pr) rather than executing arbitrary remote scripts or unverified dependencies.

forge-loop