Deft GH Arch

Explore a codebase for architectural friction, surface opportunities to deepen shallow modules, generate competing interface designs in parallel, and file a refactor RFC as a GitHub Issue.

Legend (from RFC2119): !=MUST, ~=SHOULD, ≉=SHOULD NOT, ⊗=MUST NOT, ?=MAY.

Inspired by improve-codebase-architecture from mattpocock/skills. Adapted to deft's domain model conventions and GitHub CLI workflow.

When to Use

• Codebase has grown and modules feel tightly coupled or hard to test
• New feature work keeps touching the same files unexpectedly
• User says "improve the architecture" or "this is hard to test"
• Pre-implementation cleanup before a major new feature

Prerequisites

• ! Verify gh is authenticated: gh auth status — stop and report if not

---

Core Concept: Deep Modules

A deep module (John Ousterhout, A Philosophy of Software Design) has a small interface hiding a large implementation. Deep modules are more testable, more AI-navigable, and let you test at the boundary instead of inside.

Shallow modules are the inverse: their interface is nearly as complex as the implementation. They force callers to know too much.

---

Process

Step 1: Explore the codebase

Navigate the codebase naturally. Note where you experience friction — the friction IS the signal:

• ~ Where does understanding one concept require bouncing between many small files?

• ~ Where are modules so shallow that the interface is nearly as complex as the implementation?

• ~ Where have pure functions been extracted just for testability, but real bugs hide in how they're called?

• ~ Where do tightly-coupled modules create risk at integration seams?

• ~ Which parts are untested, or hard to test?

• ! If core/glossary.md or UBIQUITOUS_LANGUAGE.md exists, read it — domain term precision matters when naming interfaces

• ⊗ Follow rigid heuristics — explore organically

Step 2: Present candidates

Present a numbered list of deepening opportunities. For each:

• Cluster — which modules/concepts are involved
• Why they're coupled — shared types, call patterns, co-ownership of a concept
• Test impact — what existing tests would be replaced by clean boundary tests

⊗ Propose interfaces at this step — present candidates only.

! Ask: "Which of these would you like to explore?"

Step 3: Frame the problem space

Before spawning sub-agents, write a brief user-facing explanation:

• The constraints any new interface must satisfy
• The dependencies it must manage
• A rough illustrative sketch to make constraints concrete (not a proposal)

Show this to the user, then immediately proceed to Step 4. The user reads while sub-agents work.

Step 4: Design multiple interfaces in parallel

Spawn 3+ sub-agents with radically different design constraints. Each produces:

1. Interface signature (types, methods, params)
2. Usage example showing how callers use it
3. What complexity it hides internally
4. How dependencies are handled
5. Trade-offs

Agent constraints:

• Agent 1: "Minimize the interface — aim for 1–3 entry points max"

• Agent 2: "Maximize flexibility — support many use cases and extension points"

• Agent 3: "Optimize for the most common caller — make the default case trivial"

• Agent 4 (if applicable): "Design around ports & adapters for cross-boundary dependencies"

• ! The sub-agent dispatch envelope MUST include templates/agent-prompt-preamble.md verbatim (or by reference) per AGENTS.md ## Multi-agent orchestration discipline (#954). The preamble carries the AGENTS.md read mandate, the #810 xBRIEF gate walkthrough, the PowerShell 5.1 non-ASCII rule, REST-over-GraphQL rules, and the mandatory DONE message protocol -- without it, the recurrence patterns documented in #954 re-fire on every fresh dispatch.

Present designs sequentially, then compare in prose. Give your own recommendation — which is strongest and why. If elements combine well, propose a hybrid. Be opinionated.

Step 5: User picks an interface

! Wait for explicit user selection (or acceptance of your recommendation) before filing.

Step 6: File the GitHub Issue

• ! Create a refactor RFC immediately using gh issue create with the template below
• ~ Issue-label hygiene: before filing, inspect the target repo's existing labels with gh label list or the labels API; choose one or more suitable existing labels when practical, or explicitly note that no label was applied. This is a recommendation, not a gate -- do not block issue creation solely because no label fits, and do not invent ad hoc labels outside the repo's existing label set.
• ! After creating, print the issue URL
• ! When the RFC is filed as an umbrella with companion child issues (e.g. an interface-extraction RFC plus one child per migrating caller), record the cohort via task slice:record -- --actor skill:gh-arch -- this is the durable production-side record consumed by task triage:audit --orphans (#1132 / D13). Same usage pattern as skills/deft-directive-gh-slice/SKILL.md Step 6; idempotent on retry. Skip when the RFC is a single-issue "file and forget" with no child cohort.
• ! Confirm skill exit: "deft-directive-gh-arch complete — RFC filed at ."
• ~ Suggest next steps: run deft-directive-refinement to slot the RFC into the xBRIEF lifecycle, or assign the issue for swarm pickup.

Issue template:

## Problem

Describe the coupling or shallowness found. What makes this hard to test or extend?
Describe modules and behaviors — NOT specific file paths or line numbers.

## Proposed Interface

The chosen interface design (from the parallel session).

Include:
- Interface signature (types / methods / params)
- Usage example
- What complexity it hides

## Dependency Strategy

How the new interface handles its dependencies (injection, ports & adapters, etc.)

## Test Impact

What existing tests this replaces, and what new boundary tests become possible.

## Trade-offs

What this design gains and gives up vs the current implementation.

## Acceptance Criteria

- [ ] Interface is implemented and passes all boundary tests
- [ ] Existing callers are migrated
- [ ] Old internal tests replaced by boundary tests
- [ ] task check passes

---

Anti-Patterns

• ⊗ Proposing an interface before presenting candidates to the user
• ⊗ Designing only one interface — parallel designs are required to surface real trade-offs
• ⊗ Filing the issue before the user approves the chosen design
• ⊗ Including specific file paths or line numbers in the issue body (couples to current layout)
• ⊗ Proceeding without gh authentication
• ⊗ Dispatch sub-agents without the canonical preamble — see templates/agent-prompt-preamble.md and AGENTS.md ## Multi-agent orchestration discipline (#954).