Wiki Log

Chronological record of all wiki operations.

[2026-04-05] init | Wiki initialized

Sources processed: 24 files (22 MD + 2 PDF)
Pages created: 21 total
- 1 overview page
- 12 concept pages
- 5 entity pages
- 3 synthesis pages
Source categories:
- Research reports: 10 (Agent Infra, Enterprise, Harness, Skills, Skill Factory)
- Academic papers: 7 (APF, Scaling Laws, SkillCraft, PAHF, CogRouter, SkillNet x2)
- Tools & case studies: 5 (Deep Agents, Memento-Skills, Uber, Security, Multi-agent swarm)
- Architecture analysis: 2 (Memory, AI Infrastructure)
Key findings:
- Industry shifted from capability to reliability competition (2025-2026)
- Multi-agent does NOT always help — sequential tasks see -70% with multi-agent
- 7B model with CogRouter outperforms GPT-4o by 40%
- Harness is the new mandatory infrastructure layer
- agentskills.io achieved 100K+ installs across 20+ platforms
- 40% of agentic AI projects may be canceled by 2027 (Gartner)

Scan results: 50 pages checked
- 🔴 Broken links: 0
- 🟡 Orphan pages: 0
- 🟠 Contradictions: 0
- 🟡 Missing concept pages: 5 identified (RAG, Planning, Reflection, Observability, LangGraph)
- 🔵 Knowledge gaps: 3 questions identified
Heal actions: Created 4 concept pages + 1 synthesis page
- RAG - Retrieval-augmented generation vs agent memory distinction
- Planning - Task decomposition, benchmarks (TaskBench, AgentBench)
- Reflection - Metacognitive self-critique, Reflexion framework
- Observability - Production monitoring, OpenTelemetry, distributed tracing
- RAG vs Memory Boundary - Architectural guidance on when to use each
Sources: All new content cross-validated from ≥2 trusted sources (arxiv.org, github.com, langchain.com, anthropic.com)

Sources processed: 14 local files + 2 web searches
Pages created: 31 total
- 1 overview page
- 8 concept pages
- 5 entity pages
- 3 synthesis pages
- 14 source pages
Source categories:
- Skills Agent research: 3 (HyperAgents, enterprise value, Skill Factory framework)
- API to CLI transformation: 4 (overview, implementation, design principles, tooling)
- China enterprise landscape: 2 (domestic players, market dynamics)
- Infrastructure: 2 (high-privilege agents, AI infrastructure comparison)
- Institutional AI: 1 (a16z analysis)
- Web research: 2 (China market 2026, cloud giants)
Key findings:
- 80% of large Chinese enterprises require private deployment (MLPS 2.0, PIPL compliance)
- Tencent WeChat integration (March 2026) gave 1B+ users agent access overnight
- Market projected to grow 75x from < $1 B (2024) t o >$ 30B (2028)
- 67% of Chinese industrial firms integrated AI into production (often government-mandated)
- CLI design achieves 10-100x token efficiency vs. MCP
- Gartner predicts 40% of enterprise agent projects will fail by 2027
- Huawei Ascend NPU provides domestic alternative to NVIDIA (export restrictions)
- agentskills.io: 20+ platforms, 100K+ installs, 500+ published skills
- Platform integration strategy: “agent-as-feature” (WeChat, DingTalk) vs. Western “agent-as-product”
Wiki health: Excellent - no structural issues, all pages well-connected

Translation scope: All 31 pages converted to bilingual format (EN + ZH)
Translation engine: GLM-5 (Zhipu AI) via Anthropic-compatible API
Processing approach: 4 batches to avoid API overload
- Batch 1: 8 concept pages (166 translation blocks)
- Batch 2: 6 entity + overview pages (84 translation blocks)
- Batch 3: 8 source pages (part 1)
- Batch 4: 9 source + synthesis pages (part 2)
Format: Each English paragraph followed by <div class="zh-trans">中文翻译</div>
Technical terms preserved: Agent, Harness, CLI, LLM, SDK, API, MLPS, PIPL, RAG, MCP, etc.
Total translation blocks added: ~450+ across all pages
Deployment: Rebuilt wiki and pushed to GitHub Pages