Delegate action-heavy tasks to Claude Code CLI through a structured bridge with routing, timeout profiles, retry handling, dead-letter logging, and concise result shaping. Use when an OpenClaw agent should hand off code changes, file operations, API/integration work, Microsoft 365 tasks, shell-oriented execution, or deep technical review to Claude while preserving a predictable contract.
Delegate suitable tasks to the bundled scripts/claude-bridge.py runner.
Use this skill when the task benefits from Claude Code as a specialist execution worker, especially for:
Do not use this skill for simple banter, tiny direct replies, or questions that can be answered from local workspace context without delegation.
Use the bridge runner directly:
python3 scripts/claude-bridge.py --json --task-type general "<task>"
Choose a more specific --task-type when possible:
code_changefile_changeapi_callm365tool_execanalysisgeneralUseful options:
--timeout-profile quick|medium|long|interactive--mode oneshot|interactive--confidence <0-1> for gray-zone local routing--queue-priority low|normal|high--max-retries <n>The bridge adds:
NEEDS_OPENCLAW signaling when an external OpenClaw action is requiredBefore delegation:
scripts/claude-bridge.py — main CLI runnerscripts/bridge_runtime.py — routing, prompt, retry, DLQ, and logging helpersscripts/test_bridge_runtime.py — runtime testsscripts/test_claude_bridge.py — bridge behavior testsRun:
python3 scripts/test_bridge_runtime.py
python3 scripts/test_claude_bridge.py
Then run a small real invocation, for example:
python3 scripts/claude-bridge.py --json --task-type general --timeout-profile quick "Say hello in one short sentence"
Keep this skill generic before publishing: