The Agent Skills Directory

[COMMAND_EXECUTION]: The framework utilizes extensive shell orchestration via scripts such as orchestrator.sh, run-workflow.sh, and gate.sh. These scripts manage the lifecycle of an autonomous run, including git operations, worktree creation, and the invocation of the primary agent CLI. These actions are fundamental to the skill's purpose of automating repository maintenance. \n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from issue bodies, which is materialised into prompts for various agent nodes (planning, implementation, and review). The skill mitigates this risk by maintaining strict context isolation between nodes, employing a multi-step pipeline where the implementer cannot see the full validation scenarios, and using deterministic shell scripts to enforce merge gates. \n- [DYNAMIC_EXECUTION]: The utility script scripts/_audit_runner.py uses the exec() function to parse a transition table from the local state.py file during machinery audits. This is performed on trusted internal configuration files to verify the integrity of the state machine. \n- [PERSISTENCE_MECHANISMS]: The install-trigger.sh tool enables the autonomous loop by creating recurring tasks via crontab (on Linux/macOS) or Windows Task Scheduler. This mechanism is explicitly documented as the method for enabling the factory's unattended mode.

build-dark-factory