# Notes / Blogs

Purpose: store non-paper signals that help position the project, especially product posts, engineering blogs, launch notes, community discussions, and articles about emerging AI coding workflows.

This folder mirrors the public board page:

- Web page: `notes-blogs.html`
- Survey bucket: `3-Paper_Survey/notes-blogs/`

## What Belongs Here

- Industry posts or product articles that motivate a trend.
- Blog examples that are close to our thesis but are not formal evidence.
- Local captures in Markdown or small HTML cards when the source is unstable or hard to reopen.
- Notes that explain how a blog relates to multimodal verification, existing-system continuation, or behavior-conditioned coding.

## What Does Not Belong Here

- Formal paper notes. Put those in `visual-feedback/` or a future topical paper folder.
- Raw benchmark data or run logs. Put those in `2-Data/` or a subproject under `0-Project/`.
- Full copyrighted article copies unless we have an explicit reason and permission. Prefer structured summaries, short excerpts, and source links.

## Entry Schema

Each blog entry should include:

- Source metadata: title, author/source, date captured, original link.
- User-facing viewpoints and project inspirations first.
- What the article claims or observes after the viewpoint section.
- Relation to our framing.
- Evidence strength and caveats.
- Project implications.
- Optional local HTML card for quick browsing.
- Source excerpts should preserve the original language when possible. Chinese sources should be recorded in Chinese; English sources should keep a short English excerpt plus a Chinese explanation.

## Folder Convention

New entries should use one folder per blog/post:

```text
notes-blogs/YYYY-MM-DD-source-short-title/
  README.md          # main survey note
  index.html         # standalone local page when useful
  x-discussion.md    # optional public discussion/comment capture
  assets/            # optional screenshots or small local assets
```

Older flat files can stay in place for link stability, but new dashboard links should prefer the folder entry.

## Current Entries

| Entry | Source Type | Main Signal | Files |
| --- | --- | --- | --- |
| PaperAgent / Loop Engineering and Claude Code primitives | WeChat article + primary-source cross-check | Coding-agent work is shifting from one-off prompts to loops that schedule, discover, build, verify, persist state, and repeat; verification/judgment becomes the scarce layer | `2026-06-27-paperagent-loop-engineering-claude-code/` |
| Anthropic harness design for long-running application development | engineering blog | Long-running coding quality comes from harness design: planner/generator/evaluator, sprint contracts, Playwright QA, and skeptical evaluation | `2026-03-24-anthropic-harness-design-long-running-apps/` |
| Kimi K2.7 Code and behavior-conditioned coding | industry article | AI coding inputs are shifting from explicit requirements to visible product behavior; useful as motivation for existing-system continuation | `2026-06-24-kimi-k2-7-code-behavior-input/` plus legacy flat files |