improve-codebase-architecture

Files (3)

• HTML-REPORT.md
• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

If the host has no Skill tool, read the named installed skill’s `SKILL.md` and relevant references directly.


# Improve Codebase Architecture

Surface architectural friction and propose **deepening opportunities**: refactors that turn shallow modules into deep ones. The aim is testability and AI-navigability.

## Invocation modes

Resolve the mode before choosing where to look. Read the invocation arguments from `$ARGUMENTS` when the host expands them, or from the user's invocation message otherwise. A leading standalone `full` argument selects the full invocation below; it is a mode keyword, not a subsystem name. Without `full`, use the default invocation. A path or phrase that merely contains the word "full" does not select full mode.

### Default invocation

Review the current project. Use recent Git history to select one frequently changed subsystem unless the user specifies a scope.

Present up to three worthwhile improvements with concrete file and call-site references, tradeoffs, testing benefits, and before/after diagrams. Generate the HTML report in the project's `docs/` folder (see [Report location](#report-location)) and open it.

Apart from adding the report file, leave repository files, tickets, and Git state unchanged. Stop after the report; begin the design interview only when requested. If no worthwhile opportunities exist, say so rather than inventing candidates.

### Full invocation

For `improve-codebase-architecture full`, apply this embedded prompt:

```text
Review the entire project codebase.

Override the default focus on one recently changed subsystem.
Map the major applications, shared packages, and their dependencies,
then examine each area for architectural friction.

Include stable and older code. Remove the three-recommendation limit
and report all worthwhile findings, ranked by impact and effort.

Include a coverage summary showing which areas were examined and
which remain unchecked. Keep the run report-only.
```

Both modes preserve existing repository files, tickets, and Git state, write the report to the project's `docs/` folder, open it, and stop. The full invocation overrides the default scope and candidate limit. It does not start the design interview or authorize implementation. If no worthwhile opportunities exist, report that finding.

## Shared design context

This command is _informed_ by the project's domain model and built on a shared design vocabulary:

- Call the Skill tool with "codebase-design" 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 these terms exactly in every suggestion, and don't drift into "component," "service," "API," or "boundary."
- The domain language in `GLOSSARY.md` gives names to good seams; ADRs in `docs/adr/` record decisions this command should not re-litigate.

## Process

### 1. Explore

**Scope before you scan: YAGNI.** Decide where to look using the selected invocation mode:

- **Full:** inventory the project's major applications, shared packages, other owned source areas, and their dependencies. Examine each area, including stable and older code. Use recent history as context without restricting coverage. Keep a ledger of area/path, dependencies, review status, and remaining work across exploration passes. Name exclusions such as generated or vendored code. If limits prevent completion, identify unchecked or blocked areas and their reasons rather than claiming a complete review.
- **Default with a named direction** (a module, subsystem, or pain point): use that scope and skip history-based inference.
- **Default without a named direction:** walk back a good stretch of the commit history (`git log --oneline`) to find the files and areas that keep coming up. Select one frequently changed subsystem for this review. If no clear hot spot emerges, choose one coherent subsystem and explain the choice in the report.

Read the project's domain glossary (`GLOSSARY.md`) and any ADRs in the area you're touching first.

Then spawn a sub-agent to walk the codebase. Don't follow rigid heuristics; explore organically and note where you experience friction:

- Where does understanding one concept require bouncing between many small modules?
- Where are modules **shallow**, with an 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 codebase 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.

### 2. Present candidates as an HTML report

Write a self-contained HTML file to the path described in [Report location](#report-location). Open it for the user (`xdg-open <path>` on Linux, `open <path>` on macOS, `start <path>` on Windows) and tell them the absolute path.

#### Report location

Always write the report into the reviewed project's `docs/` folder:

```text
<project-root>/docs/architecture-review-<project>-<MMDDYYYY>-<HHMM>.html
```

- `<project-root>`: `git rev-parse --show-toplevel`, or the current directory outside Git. Create `docs/` if it doesn't exist.
- `<project>`: the basename of `<project-root>` (for example `termight`).
- `<MMDDYYYY>-<HHMM>`: local date and 24-hour time when the report is written, from `date +%m%d%Y-%H%M` (for example `10132026-1145` at 11:45 am, `10132026-1300` at 1 pm).
- Never overwrite an existing report. If the name is taken (two runs in the same minute), append `-2`, `-3`, and so on before `.html`.

Reports must never be committed, pushed, or merged. Before writing the report in a Git repository, keep it out of Git locally:

1. Find the exclude file with `git rev-parse --git-path info/exclude` (this is correct in linked worktrees too).
2. If it doesn't already contain the line `/docs/architecture-review-*.html`, append that line, creating the file if needed. Don't edit `.gitignore` or any other tracked file.
3. After writing, confirm `git check-ignore -q <report-path>` succeeds. If `git ls-files 'docs/architecture-review-*.html'` lists any report, it is already tracked and the exclude won't hide it; tell the user rather than untracking it yourself.

Never stage or commit the report.

The report uses **Tailwind via CDN** for layout and styling, and **Mermaid via CDN** for diagrams where a graph/flow/sequence reliably communicates the structure. Mix Mermaid with hand-crafted CSS/SVG visuals: use Mermaid when relationships are graph-shaped (call graphs, dependencies, sequences), and hand-built divs/SVG when you want something more editorial (mass diagrams, cross-sections, collapse animations). Each candidate gets a **before/after visualisation**. Be visual.

For each candidate, render a card with:

- **Recommendation number**: sequential from 1 in display order; use the same number in the report and conversation so the user can select it with `implement-recommendation`
- **Files**: which files/modules are involved
- **Problem**: why the current architecture is causing friction
- **Solution**: plain English description of what would change
- **Benefits**: explained in terms of locality and leverage, and how tests would improve
- **Tradeoffs**: costs, risks, and when the change would not pay off
- **Before / After diagram**: side-by-side, custom-drawn, illustrating the shallowness and the deepening
- **Recommendation strength**: one of `Strong`, `Worth exploring`, `Speculative`, rendered as a badge

In full mode, include the project/dependency map and coverage summary from the exploration ledger. Show examined, unchecked, blocked, and excluded areas with paths and reasons where applicable. Include every worthwhile finding without the default three-candidate cap. Give each finding an impact and effort estimate with a short rationale, then rank by impact first and effort second. Assign recommendation numbers after ranking so report order and follow-up selections agree.

End the report with a **Top recommendation** section: which candidate you'd tackle first and why. If no candidate is worthwhile, explain that finding instead.

**Use GLOSSARY.md vocabulary for the domain, and the `/codebase-design` vocabulary for the architecture.** If `GLOSSARY.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. a warning callout: _"contradicts ADR-0007, but worth reopening because…"_). Don't list every theoretical refactor an ADR forbids.

See [HTML-REPORT.md](HTML-REPORT.md) for the full HTML scaffold, diagram patterns, and styling guidance.

Do NOT propose interfaces yet. After opening the report, tell the user its absolute path and stop. The user can request a design interview about a candidate in a follow-up.

### 3. Grilling loop

From corynaegle-ai/agent-skills, MIT.

More from corynaegle-ai/agent-skills