Adding OpenClaw agent by ChathurangiShyalika · Pull Request #428 · IBM/AssetOpsBench

ChathurangiShyalika · 2026-07-03T00:27:45Z

Adding OpenClaw agent [Under development]

Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

Add output field to ToolCall and register a PostToolUse hook so the MCP server response is attached to each ToolCall in the trajectory. Tool outputs are flushed onto the previous turn's calls when the next AssistantMessage or ResultMessage arrives. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

The SDK passes tool_response as a plain string, not a nested dict. The hook now handles both string and dict responses gracefully. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

feat: add ClaudeAgentRunner using claude-agent-sdk

chore(deps): update actions/stale action to v10

Use ThreadPoolExecutor to issue all (failure_mode, sensor) LLM calls concurrently instead of sequentially, reducing wall-clock time for large mapping requests. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

Versions 1.82.7 and 1.82.8 contain malicious code that exfiltrates credentials. Pin to the known-good resolved version in both pyproject.toml and benchmark/basic_requirements.txt. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

- IoT: cache asset list and sensor-per-asset lists after first CouchDB fetch - WO: cache dataset DataFrames after first load to avoid per-tool-call queries - TSFM: cache model config.json reads; read dataset once in integrated TSAD instead of re-reading per target column - FMSR: cache LLM failure-mode results to avoid duplicate calls for same asset Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

Add human-readable title to every @mcp.tool() decorator across all six servers (iot, fmsr, tsfm, utilities, wo, vibration) per MCP spec recommendation from issue #197. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

Add server-level instructions string to each FastMCP constructor so MCP clients understand the purpose of each server before inspecting individual tools. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

feat: optimize MCP servers with caching and metadata

Legacy code preserved on archive/legacy-tmp branch for reference. Removes experimental agents, sample data, and deprecated utilities from main branch to reduce clutter. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

chore: archive and remove legacy src/tmp/

New OpenAIAgentRunner under src/agent/openai_agent/ that uses the openai-agents SDK to orchestrate tool calls against existing MCP servers. - runner.py: OpenAIAgentRunner subclass with MCPServerStdio integration - models.py: Trajectory, TurnRecord, ToolCall dataclasses - cli.py: openai-agent CLI entry point with --show-trajectory, --json - tests/: 15 unit tests covering trajectory building and runner mocking Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

Document the new openai-agent CLI, Python API, environment variables, and architecture diagram alongside the existing plan-execute and claude-agent runners. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

This reverts commit 9b88fcb. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

Route through a LiteLLM proxy with --model-id litellm_proxy/<model>. Uses AsyncOpenAI client + OpenAIChatCompletionsModel + ModelProvider, mirroring the claude-agent pattern. Requires LITELLM_BASE_URL and LITELLM_API_KEY when using the litellm_proxy/ prefix. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

Document the openai-agent CLI, Python API, LiteLLM proxy support, environment variables, and architecture diagram. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

The openai-agent runner always routes through LiteLLM proxy. Default model changed to litellm_proxy/azure/gpt-5.4. Removed all references to direct OpenAI API usage and OPENAI_API_KEY. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

Keep only "How it works" and "CLI" for each agent, add a shared "Example queries" section with $query shell variable, and drop the "Connect to Claude Desktop" section. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

…257) Match the convention set by the openai-agent refactor in d25d657. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

feat: add OpenAI Agents SDK agent runner

issue #263 Signed-off-by: John D Sheehan <john.d.sheehan@ie.ibm.com>

switch back to mlflow.langchain.autolog in aobench/scenario-client

Integrates LangChain's deep-agents framework as a third agentic-loop backend alongside ClaudeAgentRunner and OpenAIAgentRunner. MCP servers are bridged to LangChain tools via langchain-mcp-adapters, and LLM calls route through the LiteLLM proxy by default. New `deep-agent` CLI entry point mirrors the existing agent CLIs. Closes #266 Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

feat: add DeepAgentRunner (LangChain deep-agents)

Adds opt-in OpenTelemetry tracing so each agent run emits a single correlated trace spanning the agent graph, every LLM call, and the LiteLLM proxy. - src/observability/ package exposes init_tracing() + agent_run_span() helpers; no-op when OTEL env not configured. - All four runners (plan-execute, claude-agent, openai-agent, deep-agent) wrap run() in a root span with gen_ai.* semconv attributes and trajectory-derived usage totals. - httpx auto-instrumentation propagates traceparent to the LiteLLM proxy so its spans nest under the agent trace. - Tests use InMemorySpanExporter; no collector required. Closes #270 Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

The three SDK runners (claude-agent, openai-agent, deep-agent) each shipped byte-identical Trajectory/TurnRecord/ToolCall dataclasses and their own _resolve_model / _LITELLM_PREFIX helpers. Consolidates into: - src/agent/models.py: canonical ToolCall, TurnRecord, Trajectory alongside the existing AgentResult. - src/agent/_litellm.py: shared LITELLM_PREFIX + resolve_model(). - Removed src/agent/{claude,openai,deep}_agent/models.py. - Collapsed six duplicated per-runner _resolve_model tests into one parametrized suite at src/agent/tests/test_litellm.py. Net -110 lines, no behaviour change. Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>

Signed-off-by: Dhaval Patel <pateldha@us.ibm.com>

Adding asset profile sample for pump