← all posts

13 june 2026

saturday

Today was a mix of workflow cleanup and pushing my “second brain” project’s UI/UX toward something I can confidently hand off (and reproduce reliably) across environments.

What I learned

  • GitHub issue auto-close behavior is stricter than I remembered: closing keywords only trigger when the PR merges into the repo’s default branch, and it’s worth being disciplined about “one concern per PR” so epics can map cleanly to child issues/PRs.
  • MCP/code-intelligence tooling has a real spectrum: tree-sitter-based symbol indexing (jcodemunch), graph-based pre-indexing for faster agent Q&A (CodeGraph), IDE-like semantic tooling via MCP backends (Serena), and “persistent context” approaches (claude-context). Reading the docs side-by-side made the tradeoffs (setup friction, token efficiency claims, language coverage) clearer.
  • A subtle but important packaging lesson for design handoffs: if your build/validator scans for components by filename conventions, “prototype copies” of components can accidentally get picked up and cause duplicate registrations. Renaming handoff artifacts to a non-scanned naming pattern and updating references is a clean fix.

What I built

  • Iterated on my minimal UI kit/design system for the 2brn prototype:
  • Standardized a stronger “active/primary” accent (red) across nav, buttons, toggles, segmented controls, and analytics visuals.
  • Improved readability with larger, fluid typography and clamp()-based spacing.
  • Tightened responsive behavior: grids collapse more gracefully, sidebar/topbar behavior is cleaner across breakpoints, and I verified it at multiple viewport widths.
  • Added/clarified interaction specs for a couple key UI patterns:
  • Converting a category selector into multi-select “pill” labels.
  • Making the timeline/hour rail interactive (scroll/jump-to-hour behavior with hover/label affordances).
  • Prepared a developer handoff package for the prototype (offline-friendly, dependency-free HTML bundle) and documented how to rebuild/verify it.
  • Also fixed the duplicate-component scan issue by renaming prototype .jsx artifacts and updating index.html/README references, then re-verified the prototype still renders correctly.

Experimenting with

  • Serena: watched the intro and brought up the local Serena dashboard to understand its “IDE for your coding agent” model (projects/tools/modes/contexts) and how it compares to built-in agent tooling.
  • React Native debugging: used React Native DevTools to chase a couple runtime warnings (navigation GO_BACK not handled, and model/tokenizer asset fetch warnings related to missing content-length headers). This helped confirm what’s “real” vs dev-only noise in the console.
  • Skimmed AI safety ecosystem resources (AISafety.com) to get a better sense of community/training options, mostly as background context for longer-term interests.