OpenClaw context bootstrap

This page defines how to keep context durable and portable for future OpenClaw sessions and new machines.

Store in Git (durable)

  • Operating philosophy and sweep cadence.
  • Failure classification standards.
  • SPARQL/query patterns that generalize.
  • Operator playbooks and runbooks.
  • API workflow contracts and tool usage patterns.

Keep local/private (do not commit)

  • API keys, bearer tokens, .env secrets.
  • Raw auth stores and private local databases.
  • Full raw transcript exports with sensitive state.

Portability principle

Same tools, any building: repo stores reusable process, while site-specific truth lives in the Open-FDD live model.

Minimal bootstrap read list for fresh clones

OpenClaw + lab (files in repo — read in order on first session):

  1. openclaw/HANDOFF_PROTOCOL.md — mailbox handoff with issues_log.md and log files.
  2. openclaw/SKILL.md — agent behavior, bootstrap modes, MCP, security scope.
  3. openclaw/references/testing_layers.md — where pytest vs bench vs bootstrap.sh live.
  4. openclaw/references/legacy_automated_testing.md — redirect from deprecated open-fdd-automated-testing if anything still points there.
  5. openclaw/references/session_status_summary.md5-bullet lab snapshot format when agents must not dump logs into provider chat.

Product and operations (published docs paths):

  1. OpenClaw integration
  2. Open-FDD integrity sweep
  3. Operator framework
  4. AI PR review playbook

AI data modeling (when the stack includes model/API): LLM workflow, AI-assisted tagging, plus GET /data-model/export and PUT /data-model/import as in OpenClaw integration.