agents-tools-and-streaming
Use AdalFlow Agent, Runner, ReActAgent, FunctionTool, ToolManager, streaming events, permissions, and MCP tools safely.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- agents-tools-and-streaming
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use AdalFlow Agent, Runner, ReActAgent, FunctionTool, ToolManager, streaming events, permissions, and MCP tools safely.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/agents-and-tools.md9.5 KB
- 参考文档references/mcp-and-permissions.md9.7 KB
- 参考文档references/streaming-and-run-results.md8.9 KB
- 参考文档references/troubleshooting.md11.6 KB
- 脚本scripts/agent_runner_fake_planner_smoke.py6.7 KB
- 脚本scripts/function_tool_smoke.py4.4 KB
03 SKILL.md 正文
Agents, Tools, and Streaming
Use this sub-skill when the task involves AdalFlow agent execution, tool wrapping, tool calling, streaming run events, human approval for tools, or MCP-served tools.
Route here for
- Wrapping Python callables as
FunctionTool, including sync, async, sync-generator, async-generator, and bound-method tools. - Inspecting or constructing
FunctionDefinition,Function,FunctionExpression,FunctionOutput, orToolOutputvalues. - Building a
ToolManager, parsing function expressions, or executing tool calls directly. - Constructing
Agent,Runner, or legacyReActAgentworkflows after the model client / generator setup is known. - Calling
Runner.call,Runner.acall, orRunner.astreamand consumingRunnerResult, step history, and stream events. - Adding human-in-the-loop approvals through
PermissionManager, CLI/FastAPI handlers,require_approval, or pre-execution confirmation details. - Integrating MCP tools through MCP server parameter dataclasses,
MCPFunctionTool, andMCPToolManager.
Route elsewhere
- Model client, provider credentials,
Generator, prompt/model kwargs, and provider streaming setup belong tomodel-client-and-generator-workflows. - DataClass schemas, parser design, and general structured output repair belong to
core-components-and-structured-iounless the schema is specifically an agent finalanswer_data_type. - Retrieval/RAG tools belong to
retrieval-rag-and-data-pipelinesbefore wrapping them for agent use here. - Tracing spans, logs, MLflow, and persistent observability belong to
tracing-observability-and-configuration. - Training and optimizer behavior for agents/tools belongs to
evaluation-and-optimization.
Read these references
- references/agents-and-tools.md for
FunctionTool,ToolManager,Agent,Runner,ReActAgent, tool outputs, and fake-planner testing. - references/streaming-and-run-results.md for
RunnerResult,StepOutput, streaming event types, and robust event consumption loops. - references/mcp-and-permissions.md for tool approval flows, permission handlers, and MCP manager/server parameter patterns.
- references/troubleshooting.md for unsafe tools, missing annotations, async-loop issues, result formatting, max-step exhaustion, finalization, permission denials, stream consumption, and missing MCP dependencies.
Safe bundled checks
python scripts/function_tool_smoke.pyverifies service-freeFunctionToolandToolManagerbehavior.python scripts/agent_runner_fake_planner_smoke.pyverifiesAgent+Runnersync, async, streaming, and permission behavior with a fake planner and no provider calls.
Run these only in an environment where adalflow is importable. They do not require API keys, network services, provider calls, or MCP servers.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).