Introduce Lecture 14 (From Single Loops to Graph Engineering) and Project 08
(Draw Your Workflow as a Graph), translated into all 15 locales. Update the
sidebar config, per-locale READMEs (what's-new blocks, learning path phases
7-8, project evolution, and outline links pointing to each locale's docs),
and the root/zh-CN READMEs.
Key content: the four stacked layers (prompt to context to loop to graph) and
where harness sits, the four parts of a graph, why in-loop checkpoints can't
fix the three structural failures at scale, a framework-agnostic six-step
build walkthrough, graph vs. workflow, anchors, and the open-source
landscape before vs. after the name.
The Goodreads URL (book/show/13563030) redirected to the wrong book.
Replace it with the Google Books link to the correct title across all
15 language editions of lecture-08 and lecture-10 (30 files total).
- Rename summary section from "带走什么"/"关键要点" to "核心要点" across
all 12 Chinese lectures for consistency
- Add bullet points and bold labels to Lecture 01 failure modes section
for better scannability
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The Chinese lecture text had a heavy "AI listing bullet points" feel —
excessive em-dashes, forced analogies on every paragraph, rhetorical
patterns like "this is not X, but rather Y", and bullet-point summaries
that read like model output. This commit rewrites the narrative sections
of all 12 Chinese lectures in a more natural, conversational tone, then
re-translates the 12 English lectures to match the polished Chinese versions.
Key changes:
- Remove repetitive analogies (restaurant ordering, carpenter, exam papers,
choir rehearsal, dorm room, etc.) — keep at most one per lecture
- Reduce em-dash usage, favor commas and periods
- Rewrite "core concepts" definitions in natural prose
- Make "key takeaways" sections feel less like model-generated summaries
- Align English versions with the rewritten Chinese structure
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Docs were restructured to add a language layer (docs/en/...) but links
in README, docs-readme locales, and projects still pointed to the old
paths (docs/lectures/, docs/projects/). Also convert three non-existent
example file references in lecture-04 to plain text.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Translate all remaining English placeholder content across ar, de, es, fr, ja, ko, ru, uz, vi, zh, zh-TW locales. Includes lecture index files, code examples, templates, references, and skills docs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Renamed 'Core Concepts' section to 'Key Terminology' to prevent structural confusion
- Replaced ambiguous 'above' reference with explicit list of the five defense layers
- Applied fixes across all language versions (en, zh, ru, vi)
- Rewrite all 12 Chinese lectures with warmer, more engaging prose
- Remove rigid "四种/五种 xxx" numbered list patterns
- Vary narrative structure across lectures instead of uniform template
- Sync English translations to match new Chinese content
- Keep mermaid diagrams, code blocks, exercises, and references intact
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Copy code/ directories from zh lectures to en lectures (12 dirs)
- Add code example GitHub links + project links to all 12 CN lectures
- Add code example GitHub links + project links to all 12 EN lectures
- Add related lecture links + template GitHub links to all 6 CN projects
- Add related lecture links + template GitHub links to all 6 EN projects
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Cover concepts that were previously missing or incomplete from the
three reference articles (OpenAI harness engineering + Anthropic
long-running agents + Anthropic harness design):
- L01: Add OpenAI's empty-repo experiment (0 human-written lines,
1M+ agent-generated lines, 1500 PRs, 3 engineers)
- L05: Deepen context anxiety analysis with Sonnet 4.5 vs Opus 4.5
data, compaction vs reset trade-offs
- L09: Add self-evaluation bias and generator-evaluator separation
with DAW experiment cost comparison table
- L10: Add layered domain architecture enforcement and custom
lint patterns from OpenAI's Codex practice
- L11: Add three-agent architecture with detailed phase timing
and cost data from Anthropic's DAW experiment
- L12: Add golden principles, garbage collection cycle, harness
simplification as models improve (Opus 4.6), and high-throughput
merge philosophy
Also fix root index.md to auto-redirect based on browser language.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
sanbuphy