subagent-dev

Automating tasks.md execution involves delegating each task to an implementer and a reviewer subagent for one-pass compliance and code-quality validation. This allows complex 1C project workstreams to progress autonomously without user coordination.

Yes, autonomous subagent execution continues through up to two correction iterations per task without user prompts. It only stops to escalate blockers when dependencies or contradictions prevent continuation.

The reviewer feedback loop works by having a dedicated reviewer subagent check both specification match and code quality after implementation. It runs up to two correction iterations autonomously before updating the task state.

You need to prepare a tasks.md file containing the atomic task steps and a design.md file. The automation requires reading both files to execute parallel subagent workflows and update task progress accurately.

Progress tracking is handled directly within tasks.md by marking completed items from [ ] to [x]. The system updates the file state automatically as the implementer and reviewer subagents finish each task successfully.

When dependencies or contradictions prevent continuation, the system stops the autonomous loop and produces a final commit-oriented report. This report escalates the specific blockers preventing further progress on the tasks.md workstream.