Quick Notes

Short, opinionated notes on AI tooling and the build itself — Claude Code, MCP, workflows, and the occasional sharp edge worth writing down.

3 days ago

Next.js

Pin a view-transition-name so it survives the prod build

React's <ViewTransition> generates a different name in dev and prod builds, so a freeze that matched in dev missed in prod. A raw inline view-transition-name is identical in both.

5 days ago

MCP

Run Codex from Claude Code over MCP

Wire OpenAI's Codex CLI into Claude Code as a second agent with four lines of .mcp.json. The argument is codex mcp-server, not codex mcp.

1 week ago

Claude Code

/deep-research is the quickest way to see a dynamic workflow

Dynamic workflows make more sense once you have watched one run. Claude Code ships with /deep-research: it fans out searches, cross-checks sources, and writes a cited report.

1 week ago

Agents

Git worktrees for parallel agent sessions

Run each agent session in its own git worktree so parallel edits never collide. Claude Code has native support: claude --worktree, or isolation: worktree on a subagent.

2 weeks ago

Next.js

Cache-Control for committed static art: never immutable

Committed, non-fingerprinted art should be cached with stale-while-revalidate, never immutable — a regenerated asset at the same path must still refresh.

2 weeks ago

Tooling

oxc: one Rust toolchain instead of ESLint + Prettier

oxlint and oxfmt replace ESLint and Prettier with one Rust toolchain — fast enough to run in a pre-commit hook without thinking about it.

3 weeks ago

Claude Code

Plan mode or a skill? When to reach for which

Plan mode and skills solve different problems. Plan mode is read-only research before a change; a skill is a procedure you have written down to reuse.

3 weeks ago

Security

Vet MCP servers before you connect them

Connecting an MCP server runs someone else's code and feeds its tool descriptions to your model. Treat it like adding a dependency with shell access — because that is what it is.