Inspect captured HTTP requests, pytest results, logs, and exceptions to diagnose application failures.
npx skills add https://github.com/smelloscope/smello --skill smello-smelloscope
Please help me install this Agent Skill. Skill: smello Source: https://github.com/smelloscope/smello/tree/main/skills/smello Command: npx skills add https://github.com/smelloscope/smello --skill smello-smelloscope
smello query CLI. - Hierarchy-aware debugging: Resolve runtime parent-child relationships via span IDs to attribute a failed request or log to the exact test or incoming request that produced it, instead of guessing from timestamps. - Instrumented run sessions: Wrap scripts, pytest suites, or web apps with smello run --app --session to capture a tagged debugging session, including gRPC calls from Google Cloud libraries. - Use Case: A user pastes a Smello dashboard URL like http://localhost:5110/#<uuid> and asks what went wrong; the Skill fetches the event, resolves its hierarchy, and explains the failure in context — for example, a ConnectError inside a specific incoming request that returned 500. ## Quick Start Ask the assistant to inspect the events captured by your Smello server at http://localhost:5110 and explain any failed requests, tests, or exceptions it finds.High-intent search queries and answers about installing and using this skill.
Run smello query with filters like --type http --status 500 --ancestors to list failing requests together with the test or endpoint that made them. Calls that raised before getting a response have a null status code, so list the session's HTTP events and look for summaries showing an exception type instead of a status.
Fetch the event with smello query <id> and inspect its hierarchy.runtime field. Search the parent_span_id (or span_id for logs and exceptions) to pull back the owning operation and its other children in one request, rather than inferring relationships from adjacent timestamps.
Yes. gRPC calls from libraries like BigQuery, Firestore, Pub/Sub, and Vertex AI appear as HTTP events with grpc:// URLs and protobuf bodies serialized as JSON. The library field in the event data shows grpc as the source.
An empty result exits 0 and means no matches, not an error. Check that the server is reachable with smello query --limit 1, that you are querying the right --app and --session, and that SMELLO_URL points at the correct server. Log capture is opt-in via SMELLO_CAPTURE_LOGS=true.
List results are capped at 200 events with no pagination, so on a busy server narrow queries with --session, --app, or --search instead of assuming the newest 200 cover the run. The smello meta command reads the whole database and is not subject to this limit.
snoodleboot-ioImplements OpenTelemetry distributed tracing with spans, context propagation, and sampling strategies.
StiensGate928Diagnose failing commands, patch root causes, and verify with narrow rechecks.
santosomarApply the SZZ algorithm to Git history to identify bug-introducing commits.
masermediagroup-stackDiagnose bugs across Python, Go, Rust, Node.js, containers, and production environments.