/export-state — State Export for Migration

You are exporting the current kernel state into a portable snapshot format. This snapshot can be imported into a standalone agent instance (or another Claude Code project) to transfer accumulated learning.

Purpose

The KernelSnapshot is the migration currency — a self-contained JSON file that preserves:

• •All events (the full learning history)
• •Current evolved rules (the accumulated expertise)
• •Prompt version history (the evolution trajectory)
• •Internal model state (world/valence/skill differentiation)

Procedure

Step 1: Gather All State

Read the following files:

1. •Event log: .claude/path-kernel/event-log.jsonl — read all lines, parse each as JSON, collect into an array.
2. •Current rules (internal models): Read .claude/path-kernel/config.json for the authoritative list of mutable rule files (agentCanModify). Read each file listed there. These span three domains under .claude/rules/: world, valence, and skill.
3. •Current CLAUDE.md: The composed mutable surface.
4. •Cycle state: .claude/path-kernel/state.json if it exists.
5. •Protected config: .claude/path-kernel/config.json.

Step 2: Assemble the Snapshot

Construct a JSON object matching the KernelSnapshot schema:

{
  "schemaVersion": 1,
  "exportedAt": "<ISO 8601 timestamp>",
  "source": "claude-code",
  "events": [<all events from event-log.jsonl>],
  "currentPrompt": "<contents of CLAUDE.md>",
  "currentVersion": <number of modification:applied events + 1>,
  "promptVersions": [<extracted from modification events>],
  "protectedConfig": <contents of .claude/path-kernel/config.json>,
  "internalModels": {
    "world": [
      { "filename": "environment.md", "content": "<file contents>" },
      { "filename": "self-model.md", "content": "<file contents>" }
    ],
    "valence": [
      { "filename": "priorities.md", "content": "<file contents>" },
      { "filename": "tradeoffs.md", "content": "<file contents>" }
    ],
    "skill": [
      { "filename": "tool-patterns.md", "content": "<file contents>" },
      { "filename": "domain-expertise.md", "content": "<file contents>" }
    ]
  }
}

Step 3: Write the Snapshot

1. •Ensure .claude/path-kernel/exports/ directory exists (create if needed).
2. •Write the snapshot to .claude/path-kernel/exports/snapshot-<timestamp>.json where <timestamp> is the current ISO date-time with colons replaced by dashes (e.g., snapshot-2026-01-15T10-30-00.json).
3. •Pretty-print the JSON (2-space indent) for human readability.

Step 4: Log the Export

Use bash to run:

bash .claude/hooks/append-event.sh "system:export" '{"file":".claude/path-kernel/exports/snapshot-<timestamp>.json"}' <version>

Step 5: Report

Tell the user:

1. •The snapshot file path
2. •Summary stats: number of events, number of rule files, current version
3. •How to use the snapshot:
   • •For another Claude Code project: Copy the snapshot into the target project's .claude/path-kernel/exports/ directory

Constraints

• •This skill is read-only with respect to kernel state — it only creates a new file in .claude/path-kernel/exports/.
• •Include ALL events — don't filter or summarize. The full history is the moat.
• •Strip the YAML frontmatter from rules files when including in internalModels (the frontmatter is Claude Code-specific metadata).