Skip to main content
Developer workflow for the agent runtime (src/agents/) in the OpenClaw repo.

​
Type checking and linting

  • Default local gate: pnpm check (typecheck, lint, policy guards)
  • Build gate: pnpm build when the change can affect build output, packaging, or lazy-loading/module boundaries
  • Full pre-push gate: pnpm build && pnpm check && pnpm check:test-types && pnpm test

​
Running Agent Runtime Tests

Run the agent runtime unit suites:
The first glob also covers the agent-tools*, agent-settings, and agent-tool-definition-adapter* suites. Live tests are excluded from the unit config; run them through the live wrapper (sets OPENCLAW_LIVE_TEST=1 and needs provider credentials):

​
Manual testing

  • Run the Gateway in dev mode (skips channel connections via OPENCLAW_SKIP_CHANNELS=1): pnpm gateway:dev
  • Trigger one agent turn through the Gateway: pnpm openclaw agent --message "Hello" --thinking low
  • Use the TUI for interactive debugging: pnpm tui
For tool call behavior, prompt for a read or exec action so you can watch tool streaming and payload handling.

​
Clean slate reset

State lives in the OpenClaw state directory: ~/.openclaw by default, or $OPENCLAW_STATE_DIR when set. Paths relative to that directory: Delete those paths for a full reset. Narrower resets:
  • Sessions only: do not delete agents/<agentId>/agent/openclaw-agent.sqlite; session rows live there alongside other per-agent state. Use /new or /reset to start a fresh session for one chat, and openclaw sessions cleanup for session maintenance.
  • Keep auth: leave agents/<agentId>/agent/openclaw-agent.sqlite and credentials/ in place.
Legacy auth-profiles.json files are no longer read at runtime; openclaw doctor --fix imports them into the SQLite store.

​
References