| name | improve-codebase-architecture |
| description | Scan a codebase for deepening opportunities, present them as a visual HTML report, then grill through whichever one you pick. |
| disable-model-invocation | true |
Improve Codebase Architecture
Surface architectural friction and pitch deepening opportunities — refactors that turn shallow modules into deep ones. The goal is testability and AI-navigability.
This command draws on the project's domain model and rests on a shared design vocabulary:
- Run the
/codebase-design skill for the architecture vocabulary (module, interface, depth, seam, adapter, leverage, locality) and its principles (the deletion test, "the interface is the test surface", "one adapter = hypothetical seam, two = real"). Use those exact terms in every suggestion — don't slide into "component," "service," "API," or "boundary."
- The domain language in
CONTEXT.md names the good seams; the ADRs in docs/adr/ record decisions this command must not re-open.
Process
1. Explore
Read the project's domain glossary (CONTEXT.md) and any ADRs covering the area you're touching first.
Then use the Agent tool with subagent_type=Explore to walk the codebase. Skip rigid heuristics — explore organically and mark wherever you feel friction:
- Where does grasping one concept mean bouncing between a swarm of tiny modules?
- Where are modules shallow — the interface nearly as complex as the implementation behind it?
- Where were pure functions carved out purely for testability, while the real bugs hide in how they're wired together (no locality)?
- Where do tightly-coupled modules bleed across their seams?
- Which parts are untested, or awkward to test through their current interface?
Apply the deletion test to anything you suspect is shallow: would deleting it concentrate complexity, or merely shuffle it elsewhere? A "yes, concentrates" is the signal you're after.
2. Present candidates as an HTML report
Write a self-contained HTML file into the OS temp directory so nothing lands in the repo. Resolve the temp dir from $TMPDIR, falling back to /tmp (or %TEMP% on Windows), and write to <tmpdir>/architecture-review-<timestamp>.html so each run gets its own file. Open it for the user — xdg-open <path> on Linux, open <path> on macOS, start <path> on Windows — and tell them the absolute path.
The report pulls Tailwind from a CDN for layout and styling, and Mermaid from a CDN for diagrams wherever a graph, flow, or sequence carries the structure reliably. Blend Mermaid with hand-built CSS/SVG visuals — Mermaid when the relationships are graph-shaped (call graphs, dependencies, sequences), hand-crafted divs/SVG when you want something more editorial (mass diagrams, cross-sections, collapse animations). Every candidate earns a before/after visualisation. Be visual.
For each candidate, render a card carrying:
- Files — which files/modules are in play
- Problem — why the current architecture creates friction
- Solution — plain-English account of what would change
- Benefits — framed in terms of locality and leverage, and how the tests improve
- Before / After diagram — side by side, custom-drawn, showing the shallowness and the deepening
- Recommendation strength — one of
Strong, Worth exploring, Speculative, shown as a badge
Close the report with a Top recommendation section: the candidate you'd take on first and why.
Use CONTEXT.md vocabulary for the domain, and the /codebase-design 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 cuts against an existing ADR, raise it only when the friction is real enough to justify reopening the ADR. Mark it plainly on the card (say, a warning callout: "contradicts ADR-0007 — but worth reopening because…"). Don't catalogue every theoretical refactor an ADR rules out.
See HTML-REPORT.md for the full HTML scaffold, diagram patterns, and styling guidance.
Do NOT propose interfaces yet. Once the file is written, ask the user: "Which of these would you like to explore?"
3. Grilling loop
Once the user picks a candidate, run the /grilling skill to walk the design tree with them — constraints, dependencies, the shape of the deepened module, what sits behind the seam, which tests survive.
Side effects happen inline as decisions firm up — run the /domain-modeling skill to keep the domain model current as you go:
- Naming a deepened module after a concept
CONTEXT.md doesn't hold? Add the term to CONTEXT.md. Create the file lazily if it's missing.
- Sharpening a fuzzy term mid-conversation? Update
CONTEXT.md on the spot.
- User rejects the candidate for 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 a future explorer would genuinely need the reason to avoid re-suggesting the same thing — skip ephemeral reasons ("not worth it right now") and self-evident ones.
- Want to explore alternative interfaces for the deepened module? Run the
/codebase-design skill and use its design-it-twice parallel sub-agent pattern.