Appendix A
Toolbox
Planned. Has an outline, but is not yet written out.
What this appendix will cover
- Current recommended tools, which change often
- Agent canvases, such as Orca, Conductor, Omnigent, Amp, Claude Code, Codex, and Grok Build
- Orchestrators, such as FirstMate by Kun Chen
- Agent workspaces, such as OpenHands
Tools I need
Test visualizer
- A visual way to quickly understand all the tests an agent proposes
- For property-based tests, show me the invariants
- For end-to-end tests, show me the spec
- For standard unit tests, just show me the test cases
- All of it nicely visualized
Design tree visualizer
- Matt Pocock's Grill Me skill is awesome
- But you don't know how many total questions there are, and it's hard to get a summary of everything that was asked
- It does generate a spec, so you could review that
- But it might be more useful to have a visual tree where the most consequential decisions are seen at the top
- Then you can keep expanding it to get a feel for whether those are worth reviewing, or whether just the big stuff at the top is good enough
Plan conformance checker
- Takes the plan I want to review, with my schemas, my system vocabulary, and my data flows
- As part of the review and the pull request, checks the implementation against them to make sure they're all in line
- If they're not, and the agent wants to depart from them, it raises the reason for doing so
- Or maybe that's actually part of validation
Layer diff viewer
- Knows the overall system architecture, and its layers
- Evaluates this particular pull request or change against it
- Essentially shows me the new layers and how they relate to the existing ones, like a layer diagram for the change
PR design explainer
- Helps me understand the design of a PR
- I don't care about the code. I care about the design, and the approach
- Focuses on the fastest possible way for me, the human, to understand what was built
- Lets me review the design and architecture, schemas, concepts, and names