Engineering Router

Use the existing AI Coding Orchestrator as the execution authority. This skill only chooses the path; it never creates a competing runtime, memory store, or evidence ledger.

Main flow

For a non-trivial change:

1. grill-with-docs when intent, terminology, scope, or acceptance criteria are still fuzzy.
2. research when an unfamiliar technology, dependency, architecture or design decision needs evidence before implementation.
3. to-spec when the work spans sessions or needs a durable build contract.
4. to-tickets when the spec decomposes into independently verifiable units with dependencies.
5. prototype when a bounded runnable experiment can resolve uncertainty faster than production implementation.
6. implement for controlled production changes. implement drives tdd where a behavioral seam exists and closes with code-review.
7. verify -> review -> artifact -> rollout using the existing orchestration and provenance contracts.
8. retro after a meaningful implementation/debugging session when the user wants environment improvements from observed evidence.

On-ramps

• Broken behavior -> diagnosing-bugs.
• Active merge/rebase conflicts -> resolving-merge-conflicts.
• Design or module-shape uncertainty -> codebase-design.
• Codebase health -> improve-codebase-architecture.
• Domain vocabulary uncertainty -> domain-modeling.
• User explicitly asks for test-first implementation -> tdd.
• User asks to review a change -> code-review.
• User asks what should improve after a session -> retro.

Rules

Keep the selected flow bounded. Carry the existing intent, context evidence, artifact, verification, and provenance identifiers through every handoff. Do not introduce a parallel planning, memory, repository graph, or receipt abstraction.

When work crosses a phase boundary, prefer continue when context is still load-bearing; otherwise use the repository's existing compaction/handoff mechanisms.

Existing ownership

• Execution authority: skills/ai-coding-orchestrator/SKILL.md
• Context authority: .ai-harness context pipeline and ContextEvidence
• Repository intelligence authority: canonical CodebaseIndex / RepositoryIntelligence
• Loop authority: .ai-harness/runtime/feedback_loop.py
• Provenance authority: existing provenance ledger
• Release authority: existing release/rollout contracts