improve-branch-architecture

Improve Branch Architecture

Surface architectural friction and propose deepening opportunities — refactors that turn shallow modules into deep ones — scoped to the current branch's changes and the radius that can reasonably fold into this branch. The aim is testability and AI-navigability.

This skill is the branch-scoped sibling of improve-codebase-architecture. It runs that skill's entire approach — the same three phases, the same vocabulary, the same deepening discipline, the same inline CONTEXT.md/ADR side effects — changed in exactly two ways:

1. Scope is the current branch's changes plus the foldable radius, not the whole codebase.
2. Medium is in-conversation markdown with ASCII before→after sketches, not a self-contained HTML report.

For a whole-codebase audit, use improve-codebase-architecture. For a correctness/bug review of the branch, use review-code.

Glossary

Use these terms exactly in every suggestion. Consistent language is the point — don't drift into "component," "service," "API," or "boundary." Full definitions in LANGUAGE.md.

• Module — anything with an interface and an implementation (function, class, package, slice).
• Interface — everything a caller must know to use the module: types, invariants, error modes, ordering, config. Not just the type signature.
• Implementation — the code inside.
• Depth — leverage at the interface: a lot of behaviour behind a small interface. Deep = high leverage. Shallow = interface nearly as complex as the implementation.
• Seam — where an interface lives; a place behaviour can be altered without editing in place. (Use this, not "boundary.")
• Adapter — a concrete thing satisfying an interface at a seam.
• Leverage — what callers get from depth.
• Locality — what maintainers get from depth: change, bugs, knowledge concentrated in one place.

Key principles (see LANGUAGE.md for the full list):

• Deletion test: imagine deleting the module. If complexity vanishes, it was a pass-through. If complexity reappears across N callers, it was earning its keep.
• The interface is the test surface.
• One adapter = hypothetical seam. Two adapters = real seam.

This skill is informed by the project's domain model. The domain language gives names to good seams; ADRs record decisions the skill should not re-litigate.

Process

1. Explore the branch's changes

Read the project's domain glossary (CONTEXT.md, if any) and any ADRs in docs/adr/ for the area you're touching first, so recommendations use the project's domain language and do not re-litigate recorded decisions.

Resolve branch scope exactly as review-code does:

1. Resolve the repository default branch with gh repo view --json defaultBranchRef --jq .defaultBranchRef.name or git rev-parse --abbrev-ref origin/HEAD, stripping any leading origin/.
2. Compute the review base with git merge-base origin/<default-branch> HEAD.
3. The change set is the committed diff from that merge-base to HEAD, plus staged, unstaged, and untracked changes. Include deleted files.

Then use the Agent tool with subagent_type=Explore to walk the change set and gather context. The exploration is bounded by the branch, but you must read past the diff hunks to judge architecture honestly:

• Read each changed file in full, not just the changed lines — depth and seams aren't visible from hunks alone.
• Read the unchanged seam-partners of the changed code: the callers and callees the changed code interfaces with, so you can judge the interface and the deletion test against real usage.

Degrade to running inline if subagents are unavailable. This is an advisory skill, not a hard gate — do not halt when no Explore subagent surface exists.

Explore organically and note where you experience friction:

• Where does understanding one concept require bouncing between many small modules?
• Where are modules shallow — interface nearly as complex as the implementation?
• Where have pure functions been extracted just for testability, but the real bugs hide in how they're called (no locality)?
• Where do tightly-coupled modules leak across their seams?
• Which parts of the change are untested, or hard to test through their current interface?

Apply the deletion test to anything you suspect is shallow: would deleting it concentrate complexity, or just move it? A "yes, concentrates" is the signal you want.

The candidate filter is "anything foldable into this branch." Surface friction whose fix could plausibly ride along in this branch:

• Friction the branch introduced.
• Pre-existing shallowness in the files the branch touches.
• Foldable issues in unchanged neighbors (callers/callees), as long as the change set stays reasonable.

Bound recommendations by keeping the change set reasonable — do not hand the contributor a sprawling refactor of unrelated, unchanged code. That is the job of improve-codebase-architecture.

2. Present candidates as in-conversation markdown

Present candidates directly in the conversation as markdown cards. Write no HTML file and open nothing. Markdown renders in any surface, including a terminal where HTML and Mermaid do not.

For each candidate, a card with these fields:

• Files — which changed files/modules (and which foldable neighbors) are involved.
• Problem — why the current architecture causes friction, in the LANGUAGE.md vocabulary, with deletion-test reasoning.
• Solution — plain-English description of what would change.
• Benefits — explained in terms of locality and leverage, and how the test surface improves, naming the dependency category from DEEPENING.md.
• Before → After sketch — an ASCII sketch (not a fenced Mermaid/HTML diagram) illustrating the shallowness and the deepening. Use boxes, arrows, and indentation so the shallow→deep shift is visible in any surface.
• Strength — a badge, one of Strong, Worth exploring, Speculative.

End with a Top recommendation: which candidate to tackle first and why.

Use CONTEXT.md vocabulary for the domain, and LANGUAGE.md vocabulary for the architecture. If CONTEXT.md defines "Order," talk about "the Order intake module" — not "the FooBarHandler," and not "the Order service."

ADR conflicts: if a candidate contradicts an existing ADR, only surface it when the friction is real enough to warrant revisiting the ADR. Mark it clearly in the card (e.g. "contradicts ADR-0007 — but worth reopening because…"). Don't list every theoretical refactor an ADR forbids.

Do NOT propose interfaces yet. After presenting the cards, ask the user: "Which of these would you like to explore?"

3. Grilling loop

Once the user picks a candidate, drop into a grilling conversation. Walk the design tree with them — constraints, dependencies, the shape of the deepened module, what sits behind the seam, what tests survive.

Classify the candidate's dependencies (in-process, local-substitutable, remote-but-owned, true-external) and recommend the matching seam/adapter/testing strategy so the deepened module is testable through its interface. See DEEPENING.md.

Side effects happen inline as decisions crystallize:

• Naming a deepened module after a concept not in CONTEXT.md? Add the term to CONTEXT.md using CONTEXT-FORMAT.md. Create the file lazily if it doesn't exist.
• Sharpening a fuzzy term during the conversation? Update CONTEXT.md right there.
• User rejects the candidate with a load-bearing reason? Offer an ADR, framed as: "Want me to record this as an ADR so future architecture reviews don't re-suggest it?" Only offer when the reason would actually be needed by a future explorer to avoid re-suggesting the same thing — skip ephemeral reasons ("not worth it right now") and self-evident ones. See ADR-FORMAT.md.
  • When the repo documents its own ADR file-naming scheme (this repo names ADRs after the originating issue — see docs/adr/README.md), follow the repo over the default in ADR-FORMAT.md.
• Want to explore alternative interfaces for the deepened module? Use the parallel sub-agent interface-design pass in INTERFACE-DESIGN.md.

Attribution

This skill mirrors the improve-codebase-architecture and grill-with-docs skills in mattpocock/skills. Its bundled reference files are copied verbatim from those skills:

• LANGUAGE.md, DEEPENING.md, and INTERFACE-DESIGN.md from improve-codebase-architecture.
• CONTEXT-FORMAT.md and ADR-FORMAT.md from grill-with-docs.

The parent's HTML-REPORT.md is intentionally not copied — this skill emits in-conversation markdown, never HTML.