feat: quality blitz — accuracy audit 203 tutorials + 3 new deep tutorials

[johnxie]

Summary

Three new deep tutorials (27,616 lines of new content):

• autoresearch (karpathy/autoresearch, 71K ⭐): MuonAdamW optimizer with Polar Express orthogonalization, fixed 300s wall-clock budget, val_bpb metric, autonomous overnight ML experiments
• hermes-agent (NousResearch, 66K ⭐): 3-layer memory (FTS5 SQLite + semantic MEMORY.md + SKILL.md procedural), 20+ platform gateways, RL trajectory → Atropos closed learning loop
• autoagent (HKUDS/MetaChain, 9K ⭐): context_variables stripping pattern, DockerEnv TCP server, XML-as-IR agent creation, self-modifying pipelines, GAIA benchmark evaluation

Critical rewrites (fabricated/wrong-repo content replaced):

• deer-flow: 8 chapters describing Apache Airflow-style DAGs → real LangGraph StateGraph + 14-stage middleware pipeline from bytedance/deer-flow
• anthropic-skills: was targeting wrong repo (easing functions + GIF utilities) → correctly covers anthropics/anthropic-quickstarts
• goose: all 8 chapters cited diagnostics-viewer.py for every topic → accurate Rust crate citations (goose-cli/src/cli.rs, session/mod.rs, goose-mcp/src/memory/mod.rs, etc.)

Accuracy audit across all 203 tutorials (~1,600+ chapters reviewed):

• ~500+ wrong source citations replaced (CI bots, release scripts, UI marketing components cited for every chapter regardless of topic)
• DSPy API: v1 (dspy-ai / dspy.OpenAI) → v2 (dspy / dspy.LM)
• Haystack: 1.x import paths → 2.x paths
• outlines org: outlines-dev → dottxt-ai everywhere
• firecrawl org: firecrawl → mendableai
• phidata → Agno rebranding note + repo pointer
• ~300+ missing mermaid diagrams added across 50+ tutorials
• Archived-project notices added to superagi, sweep
• Star counts corrected: opencode (133k→142k), openspec (35.5k→39.3k)

openclaw patched with 3 missing subsystems: ACP layer (extensions/acpx/), memory-host-sdk (8 embedding backends + QMD), Plugin SDK (plugin-package-contract interface + CLI)

README & discoverability:

• Star history chart embedded for FOMO/social proof
• Back-to-top nav on every major section
• Catalog updated: 203 tutorials, 1,624+ chapters
• Topics updated: removed low-traffic, added generative-ai, ai-tutorial
• Catalog scripts regenerated: 1,830 markdown files, 706,049 lines

Test plan

• All 3 new tutorial directories render correctly on GitHub
• Catalog manifest (tutorials/tutorial-manifest.json) reports 203 tutorials
• README star history badge loads
• deer-flow chapters cite real bytedance/deer-flow files
• goose chapters cite real Rust crate files
• DSPy code examples use v2 API (dspy.LM(...))
• CI passes (link checks, frontmatter validation)

[johnxie]

Code review

Found 3 issues (all fixed in b1c9770):

1. Raw TODO template placeholders left in published chapter — [TODO: Replace with the first main section] and [TODO: Add content here] were committed verbatim instead of being replaced with actual content.

https://github.com/johnxie/awesome-code-docs/blob/5884ab3bbbbe2e1c5e7a55da2b1157e7d2803a63/tutorials/awesome-claude-skills-tutorial/08-team-adoption-and-ongoing-maintenance.md#L52-L60

2. "See all 30+ MCP tutorials" link pointed to a single tutorial (tutorials/mcp-python-sdk-tutorial/) instead of an index or section anchor. A reader clicking it would land in one tutorial with no way to discover the other 30+.

https://github.com/johnxie/awesome-code-docs/blob/5884ab3bbbbe2e1c5e7a55da2b1157e7d2803a63/README.md#L73-L75

3. Stale README statistics: verification date still said 2026-03-20 (23 days old despite major audit on 2026-04-12), and markdown line count showed 1,048,763 instead of actual 706,049 (off by 32%).

https://github.com/johnxie/awesome-code-docs/blob/5884ab3bbbbe2e1c5e7a55da2b1157e7d2803a63/README.md#L101-L103
https://github.com/johnxie/awesome-code-docs/blob/5884ab3bbbbe2e1c5e7a55da2b1157e7d2803a63/README.md#L616-L618

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}