nightly

Automating continuous coding tasks from a GitHub issue backlog is handled by a priority-cascade selection mechanism that picks in-flight work, approved RFCs, or ranked issues to execute implementation, testing, and PR landing autonomously.

An autonomous coding agent handles task isolation by creating config-aware worktrees, ensuring changes are implemented safely and reproducibly without conflicting with the main repository branches during continuous delivery execution.

Running a self-directed coding agent without user clarifications requires an absolute autonomy contract that never blocks for input, instead recording uncertainties to disk-based artifacts and routing refusals through approval files for later review.

Auditing and replaying an automated continuous delivery coding session is achieved by persisting run artifacts to a dedicated runs directory, capturing session event logs, status transitions, and per-task plans for comprehensive later review.

Delegating implementation and testing roles within a coding agent session is best done via session forking, which splits the opencode session into specialist roles like implementer, tester, reviewer, and researcher to collect isolated results.

Opencode supports continuous delivery workflows by enabling session forking for specialist delegation and SSE event mirroring, allowing isolated execution streams and real-time event synchronization throughout the autonomous coding lifecycle.