Skip to content
Go back

Full-stack learning: get one path working end to end

约 1 分钟

Full-stack study scatters easily: frameworks, languages, databases, deploy, auth. Each rabbit hole is deep. What I care about now is simpler: get one complete path working.

A minimal loop

For example:

User opens a page → request goes out → server handles it → data is read or written → response returns → UI updates

If that path works, swapping the middle pieces later is less scary. If you only stack tutorials and never wire it yourself, integration bugs feel like fog.

What I deliberately write down

  • Where the boundary is: what the client owns vs the server — don’t smear them together
  • How failure looks: timeouts, empty data, auth errors — what the page shows and what the logs say
  • What to reuse next time: folder layout, API conventions, how env vars are managed

Agent projects are the same story. Models look flashy; the hard part is keeping tools, state, and edge cases steady.

Ideas belong here too

Besides study notes, I keep product / tech ideas — even immature ones. Publishing them forces plain language, and makes restarting easier later.

If you’re learning full-stack or building agents, use these notes as a mirror, not a syllabus.


Share this post:

接着读

  1. Scroll storytelling: smooth scroll, pin, parallax, progressA zero-dependency demo of four scroll storytelling techniques: smooth scrolling, pinned horizontal travel, parallax layers, and a scroll-driven counter, with a GSAP and Lenis counterpart.
  2. Running 24,000 particles with raw WebGLGenerate the initial particle data once, then calculate each frame's positions in a vertex shader. The complete interactive demo runs inside the post.
  3. What changed in the MCP draft: sessions and handshake are goneNotes from the official MCP spec draft's changelog, compared against the 2025-11-25 version, focused on sessions, the handshake, MRTR, and migration impact.

Discussion

评论

用 GitHub 账号登录即可留言。想法、纠错、补充都欢迎。