[FEATURE] Export session context as transferable knowledge artifact for cross-session continuity

Problem

When working on complex multi-session projects, each new Claude Code session starts from zero context — even when prior sessions have already built deep understanding of the codebase, made architectural decisions, and identified patterns.

Current workarounds are insufficient:

| Mechanism | What it preserves | What's lost |
|-----------|------------------|-------------|
| CLAUDE.md | Static instructions | Dynamic understanding, decision rationale |
| Auto memory | Factual snippets | Reasoning chains, architectural context |
| --continue / --resume | Full conversation | Not transferable to new contexts or agents |
| Subagent prompts | What you explicitly pass | Everything you don't think to include |

Real-world example

In a single session, Claude Code:

1. Ran a full-stack audit (architecture, security, AI/ML, testing) across 20+ files
2. Built deep understanding of an agent system (tools, memory, prompts, orchestration)
3. Made architectural decisions (agentic search patterns, document indexing strategy)
4. Identified 53 findings across 6 domains

To continue this work in a new session, the user must either:

• Resume the exact session (fragile, not transferable across machines)
• Manually write summaries into CLAUDE.md/memory (lossy, tedious)
• Re-explain everything from scratch (wastes tokens and time)

Proposed Solution

context export — Export accumulated session knowledge as a structured artifact

# Export current session's accumulated context
claude context export --output .claude/contexts/agent-audit-2025.md

# Import in a new session
claude context import .claude/contexts/agent-audit-2025.md

# Or reference inline
claude --context .claude/contexts/agent-audit-2025.md

The exported artifact would include:

• Codebase map: files read, their roles, relationships discovered
• Decisions made: what was decided and why (not just what)
• Findings/analysis: structured output from research/audit work
• Domain model: entities, patterns, and conventions understood
• Open threads: unfinished work, next steps, blockers

Key differences from existing features

| Feature | context export |
|---------|-----------------|
| Memory | Facts only → Full reasoning context |
| CLAUDE.md | Static, manual → Dynamic, auto-generated |
| --resume | Same session only → Transferable artifact |
| Subagent prompt | One-shot, manual → Structured, reusable |

Use cases

1. Session handoff: Morning session builds context → afternoon session continues seamlessly
2. Cross-machine: Export from laptop → import on CI/server agent
3. Team sharing: One developer's deep-dive exported for another to build on
4. Agent SDK: Parent agent exports context → child agents import relevant slices
5. Audit/review workflows: Audit findings as importable context for the fix session

Related issues

• #24798 — Inter-session communication (parallel coordination)
• #12790 — Subagent context inheritance (parent → child)
• #35238 — Session chaining (sequential continuity)
• #33727 — Project as container for sessions

This proposal complements all of the above by providing a portable, structured context artifact rather than relying on session identity or live communication channels.