tracing-observability
Configure OpenAI Agents Python tracing, trace processors, spans, sensitive-data controls, logging, usage tracking, and agent visualization.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- tracing-observability
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Configure OpenAI Agents Python tracing, trace processors, spans, sensitive-data controls, logging, usage tracking, and agent visualization.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/tracing-reference.md9.9 KB
- 参考文档references/troubleshooting.md6.9 KB
- 参考文档references/workflows.md7.8 KB
- 脚本scripts/check_tracing_config.py10.8 KB
03 SKILL.md 正文
Tracing & Observability
Use this sub-skill when a task mentions tracing, trace_include_sensitive_data, workflow_name, trace_id, group_id, trace_metadata, Span, trace processors, usage, debug logs, visualization, Graphviz, or a tracing/export API key.
Start Here
- Read references/tracing-reference.md for verified tracing APIs, config fields, processors, spans, sensitive-data controls, usage, logging, and visualization surfaces.
- Use references/workflows.md for implementation patterns: enable/disable tracing, custom processors, trace metadata, usage collection, visualization generation, and safe logging.
- Use references/troubleshooting.md for missing trace API keys, disabled tracing, sensitive payload leakage, processor/exporter failures, flush/shutdown behavior, and Graphviz optional dependency issues.
- Run scripts/check_tracing_config.py for a safe no-network import/config check that masks secret values and can emit JSON.
Routing Boundaries
- Stay here for
RunConfig.tracing_disabled,RunConfig.tracing,RunConfig.trace_include_sensitive_data,workflow_name,trace_id,group_id,trace_metadata,trace(), span helpers,TracingProcessor,add_trace_processor(),set_trace_processors(),flush_traces(),Usage, SDK loggers, anddraw_graph(). - Route model/provider API key selection, OpenAI-compatible endpoints, Responses websocket transport, and adapter-specific model setup to ../models-providers/SKILL.md.
- Route core
Runnercall shapes,RunResult, streaming event interpretation, sessions, andRunStateresume mechanics to ../core-runtime/SKILL.md. - Route realtime and voice-specific tracing controls, including audio payload controls, to ../realtime-voice/SKILL.md.
Fast Decisions
- Tracing is enabled by default; disable it globally with
OPENAI_AGENTS_DISABLE_TRACING=1orset_tracing_disabled(True), and per run withRunConfig(tracing_disabled=True). - Set a logical
workflow_namefor every production workflow and usegroup_idto link traces from one conversation or job family. - Set
trace_include_sensitive_data=Falsefor sensitive workloads; it redacts model/tool trace payloads but does not automatically sanitize application exceptions, logs, metadata, or custom processor output. - Use
add_trace_processor()to observe traces in addition to the default OpenAI exporter; useset_trace_processors()only when intentionally replacing default export behavior. - Call
flush_traces()after a trace context closes when a short-lived job or background task needs immediate export delivery. - Use
result.context_wrapper.usagefor token/request accounting anddraw_graph()for local topology visualization when the optional Graphviz package is available.
Safety Notes
- Never print API key values, per-run
tracing={"api_key": ...}values, or local machine paths in diagnostics. - Keep custom processors non-blocking and defensive; processor/exporter exceptions are logged as non-fatal but can drop observability data.
- Treat verbose logs and exception chains as separate leakage channels from tracing payload controls.
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).