Skip to content

comet-ml/opik-skills / skills/opik

Reference for the Opik SDK — tracing, span types, framework integrations, threads, and the prompt library (Python, TypeScript, REST). Use for "what span types exist", "how do I flush", "track_openai", "add OpikTracer", "version a prompt". To instrument a repo end to end, use the `opik-instrument` skill.

references/observability.md

View on GitHub
# LLM Observability Core Concepts

Keep this mental model in mind when integrating Opik into an agent or LLM app.

## Trace Boundaries

A trace should represent one complete interaction or workflow run.

Good defaults:

- one user message -> one agent response
- one API request -> one API response
- one workflow execution -> one trace

Avoid:

- one trace across multiple chat turns
- unrelated work in the same trace
- overly granular traces when spans are enough

## Spans

Spans are the steps inside a trace. Use them to break the run into meaningful operations.

Valid span types:

- `general`: orchestration or custom logic
- `llm`: model calls
- `tool`: tools, retrieval, external calls
- `guardrail`: validation or safety checks

Integration rule:

- use descriptive names
- set the right span type
- end spans when the operation completes

## Threads

Use `thread_id` to group multiple traces into one conversation or multi-step workflow.

Use threads for:

- chat agents
- support conversations
- workflows that continue across multiple user interactions

Do not use one trace for multiple turns. Use one trace per turn and share the same `thread_id`.

For concrete usage, see `SKILL.md` and `references/tracing-python.md`.

## Metadata and Tags

Use metadata and tags to make traces filterable and comparable later.

Good metadata examples:

- user or session identifiers
- environment
- experiment or rollout variant
- model version
- business context

Good tag examples:

- `production`
- `customer-support`
- `rag`
- `needs-review`

## Feedback Scores

Feedback scores are useful when you want to attach quality signals to traces.

Common sources:

- user feedback
- human review
- automated evaluation
- heuristics or post-processing

For concrete trace update patterns, see `references/tracing-python.md`.

## Distributed Tracing

If one request crosses service boundaries, propagate trace headers so downstream work stays connected to the same overall trace.

For concrete propagation examples with `get_distributed_trace_headers()` and `distributed_headers`, see `references/tracing-python.md`.

## Sensitive Data

Be careful about what you log.

- do not log secrets or credentials
- avoid raw PII unless you explicitly need it
- prefer anonymization or selective logging for sensitive data
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall