feat: add search-retrieval, memory-protocol, and obsidian-vault skills

[alexeyaudi]

New Skills from Gemini CLI Agent System

Three new skills extracted and generalized from a production Gemini CLI agent setup.

🔍 search-retrieval

Intelligent 4-tool decision tree for external data tasks:

• Google Grounding → facts, news synthesis
• SearXNG → structured search with URLs (YouTube, science, news)
• Gemini fetch → read a specific URL
• Firecrawl → JS/SPA scraping (use sparingly, 500-req limit)

Includes anti-hallucination rules for links and a critical HTTP encoding note (Python requests over curl).

🧠 memory-protocol

Persistent agent memory via a structured Markdown file:

• Read-before-answer / write-after-task cycle
• Structured entry format with scope + expiry
• /dream consolidation command for files > 200 lines
• Replace-not-append rule to prevent duplicates

📚 obsidian-vault

Two-layer knowledge architecture for Obsidian + AI agents:

• Raw Layer (Inbox, Reports, etc.) — type: raw_material, agent write zone
• Wiki Layer — read-only for agents, written only by Wiki Compiler
• Three-stage pipeline: Librarian → Wiki Compiler → Semantic Linker
• Frontmatter format, tag rules, file routing from Inbox

Each skill follows the existing README.md + SKILL.md format from skills/gemini/.