executing-plans
executing-plans
Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available
tracing-and-instrumentationEmit, inspect, adapt, and troubleshoot Agent Lightning spans, rewards, operation traces, OpenTelemetry/AgentOps tracers, and token-ID signals.
Independent GitHub snapshot · Automated classification · Original documentation governs
Use this sub-skill when a task is about spans, rewards, trace collection, adapters, OpenTelemetry/AgentOps/Weave tracer choices, or missing token/logprob signals.
| Request | Read/run |
|---|---|
| Emit rewards, messages, objects, exceptions, or operation spans | references/tracing-workflows.md |
| Inspect final reward or convert traces for algorithms | references/api-reference.md |
| Debug no spans, no active tracer, missing reward, disconnected spans, token IDs | references/troubleshooting.md |
| Run a local no-service trace smoke | python scripts/local_trace_smoke.py |
| Start stores/proxies/metrics services | route to ../cli-and-services/SKILL.md |
OtelTracer is the minimal local tracer for explicit OpenTelemetry and emitter tests.AgentOpsTracer is the default trainer tracer and instruments many LLM/agent frameworks locally.propagate=False is used for offline local span creation.find_final_reward for the last reward value.import agentlightning as agl
tracer = agl.OtelTracer()
store = agl.InMemoryLightningStore()
rollout = await store.start_rollout(input={"origin": "debug"})
with tracer.lifespan(store):
async with tracer.trace_context(
"debug-trace",
store=store,
rollout_id=rollout.rollout_id,
attempt_id=rollout.attempt.attempt_id,
):
agl.emit_message("hello")
agl.emit_reward(1.0)
spans = await store.query_spans(rollout.rollout_id)
assert agl.find_final_reward(spans) == 1.0
The bundled scripts/local_trace_smoke.py runs this with extra operation, tag, link, and object assertions.
This sub-skill owns spans and adapters. It does not own agent function signatures, store status transitions, or service startup. Route those to agent-authoring, runner-store-training, and cli-and-services.