
Execute a bounded coding task in the Autoflow harness. Use when a specific task has a defined scope, acceptance criteria, and repository context and needs code changes from Codex, Claude Code, or another coding CLI.
Orchestrate the Autoflow control loop from OpenClaw or another outer agent host. Use when the system must inspect workflow state, choose the next role, dispatch a background run, and close the loop after execution.
Implements the closed-loop autonomous development cycle (discover→fix→document→code→test→refactor→commit) for rapid 1-2 minute iteration cycles
Run Autoflow as a continuous delivery loop. Use when the system should inspect workflow state, commit dirty changes, push to GitHub, and dispatch the next ready role on a schedule.
Perform recurring maintenance inside Autoflow, including backlog grooming, dependency updates, flaky test cleanup, and low-risk repository hygiene. Use for scheduled upkeep or after autonomous delivery cycles.
Review Autoflow-generated changes for bugs, regressions, missing tests, and acceptance-criteria gaps. Use after any implementation run and before a task is marked complete or committed.
Create or update product specs for Autoflow-style autonomous development. Use when the system needs a scoped implementation spec, interfaces, constraints, risks, or acceptance criteria before task decomposition or coding.
Build and maintain the task graph for Autoflow. Use when a spec must be decomposed into executable tasks, dependencies, priorities, and acceptance criteria in a stable machine-readable format.