Subagent-Driven Development

Overview

Execute implementation with focused subagent handoffs in one session and minimal orchestration overhead.

Use this only after scope and plan are explicit.

Default sequence

1. •TaskPlanner: refine implementation steps
2. •Coder: apply scoped changes
3. •Reviewer pass 1: validate scope/spec fit
4. •Reviewer pass 2: validate quality and residual risk

Execution rules

• •Provide full task context to each subagent.
• •Use @.opencode/context/project/subagent-handoff-template.md for every delegation.
• •Keep one active implementation stream unless tasks are independent.
• •Re-run both review passes after non-trivial fixes.
• •Stop when scope is satisfied; avoid extra "nice to have" additions.

If multiple tasks are independent, batch them and use parallel delegation.

Reviewer focus

• •Confirm requested behavior is present.
• •Confirm no unrequested behavior was added.
• •Confirm project runtime preferences were respected.
• •Return clear pass/fail with concise findings.

Pass 1 should fail fast on scope drift. Pass 2 should prioritize risk and quality issues by severity.

Guardrails

• •No worktrees unless explicitly requested.
• •No automatic git actions unless explicitly requested.
• •Tests are optional unless user/policy requires them.

Escalation path

• •Requirement ambiguity -> return question to orchestrator before coding.
• •Scope conflict -> prioritize explicit user instruction.
• •Missing context -> call ContextScout before proceeding.

When to use this vs core delegation

• •Use hf-subagent-driven-development when planning is complete and the remaining work is execution.
• •Use hf-core-delegation when discovery, routing, and planning still need orchestration.