Move from Claude Code to OpenClaw without losing the coding state.
Carry the current diff, decisions, failed attempts, checks, and next action into OpenClaw without pretending that a raw Claude Code transcript is project truth.
What actually needs to move.
Repository facts
Branch, changed and untracked files, recent commits, dependency impact, and exact command or test results.
Task context
The goal, decisions, constraints, rejected approaches, unresolved questions, and the smallest useful next action.
Claude context
Persisted conversation details that explain intent. Claude Code also supports /export, local sessions, CLAUDE.md, and auto memory.
OpenClaw context
A bounded handoff plus the project itself. OpenClaw keeps durable memory in workspace Markdown and stores sessions separately from the workspace.
Claude Code to OpenClaw handoff workflow.
git status --short, git diff --stat, and the smallest relevant check. Use /export when the visible conversation contains a decision that is not represented on disk..openclaw/context.md. Start OpenClaw in the project and have it read the packet before making changes.MEMORY.md as curated long-term memory and dated files as working notes.Do not turn the transcript into memory.
Claude Code stores CLI sessions locally and can export a readable transcript. That transcript is useful evidence of intent, but it may contain guesses, superseded plans, failed tool calls, and claims that were never checked. OpenClaw likewise says its durable MEMORY.md is not a raw transcript or exhaustive archive. The handoff should distill what the next tool needs and point back to verifiable files.
A compact handoff OpenClaw can verify.
Goal:
Current branch:
Files changed by Claude Code:
Verified checks and results:
Decisions and constraints:
Failed approaches:
Unresolved risks:
Next safe action:
Verification command:Keep this packet small enough to inspect. Attach exact paths and commands instead of pasting thousands of lines of conversation history.
How ShardStitch handles the transfer.
ShardStitch can recover locally persisted Claude Code context, combine it with Git and disk evidence, and format the result for OpenClaw. The OpenClaw handoff is local, and the next agent receives explicit trust labels so it can distinguish a verified file fact from a previous model's explanation.
Included
Useful persisted context, changed files, repository state, decisions, failed checks, risks, and the next verification step.
Excluded
Secrets, raw credentials, irrelevant transcript history, and output that never persisted locally.
Result
OpenClaw begins from current project truth and one bounded action instead of rediscovering the entire task.
Sources and verification.
- Claude Code docs: sessions, resume, branch, export, and local transcripts
- Claude Code docs: CLAUDE.md and auto memory
- OpenClaw docs: workspace, instructions, memory, and sessions
- OpenClaw docs: curated memory and dated working notes
FAQ.
Can I move a Claude Code session to OpenClaw?
You can move the useful working state. Preserve the repository, extract only relevant session context, and give OpenClaw a verified continuation packet rather than expecting a direct vendor-to-vendor chat import.
Should I paste the whole Claude transcript into OpenClaw memory?
No. Keep durable memory concise and stable. Put detailed session history in a separate working record and let OpenClaw retrieve only what the current task needs.
Does the handoff upload my code?
No. ShardStitch's normal recovery path is local-first. The packet is generated from local session artifacts and repository state.
Scope note: This is a state handoff, not an account, billing, model, or permission migration. Recheck tool access and project permissions in OpenClaw before execution.