Add 'know context' command for semantic session context loading

[jordanpartridge]

Problem

Current session hooks use static CLAUDE.md files for project context. This means:

• Context gets stale over time
• No semantic relevance - just dumps entire file
• Can't auto-discover patterns from knowledge base
• No ranking by usage/recency/confidence

Proposed Solution

Add ./know context command that semantically searches Qdrant for relevant project context.

Features

1. Auto-detect project from git repo or directory name
2. Smart search for:
   • Recent decisions (updated_at DESC)
   • High-usage patterns (usage_count DESC)
   • Project-specific architecture (module=<project>)
   • Common gotchas (category=gotcha)
3. Rank by relevance:
   • Recently accessed = still relevant
   • High confidence = vetted knowledge
   • Frequently used = proven patterns
4. Format for Claude:
   • Concise summaries with links to full details
   • Grouped by category (architecture, patterns, decisions, gotchas)
   • Token-aware (max-tokens flag to fit context window)

Usage

# Auto-detect project and generate context
./know context

# Specify project explicitly
./know context --project=conduit-ui

# Limit output tokens for session hooks
./know context --max-tokens=2000

# Filter by specific categories
./know context --categories=architecture,patterns

Session Hook Integration

#!/bin/bash
# ~/.claude/hooks/session-start.sh
./know context --project="$(basename "$PWD")" --max-tokens=2000

Implementation Notes

• Use QdrantService search with filters
• Group results by category
• Format as markdown with summary sections
• Track which entries are used (increment usage_count)
• Consider caching recent context lookups

Acceptance Criteria

• ./know context command exists and works
• Auto-detects project from pwd/git
• Searches Qdrant with smart ranking
• Formats output for Claude consumption
• Respects --max-tokens limit
• Increments usage_count for accessed entries
• Has comprehensive tests
• Documentation in README

Benefits

• Dynamic context that evolves with your knowledge base
• Semantic relevance instead of static dumps
• Token-efficient - only includes what matters
• Self-improving - tracks what context is actually useful (usage_count)
• Multi-project - works across all your projects automatically