evrentan/observe-lens

Read-only investigation across CI pipelines, CD deployments, databases and telemetry (logs, traces, metrics, alerts) to explain what happened to a build, a deploy, a request, an order, a test run or a feature, and why. Routes to observe-lens-pipelines (Jenkins, GitHub Actions), observe-lens-cd (Argo CD, Flux, Helm, Kubernetes rollouts), observe-lens-db (PostgreSQL, MySQL, MongoDB) and observe-lens-telemetry (Grafana, Loki, Tempo, Prometheus), and correlates their evidence on one timeline. Use this whenever a developer, tester, analyst or product owner asks why something failed, broke, is slow or behaves differently, whether a change is deployed or verified, what a service logged, what state a record is in, or wants a status or root-cause summary across systems, even if they only paste an id, a URL, an error message or a time window and do not name a tool. Also use it to set up or check observe-lens configuration. Never changes the systems it reads.

Read continuous-delivery state read-only to explain what is deployed where and whether it is healthy: Argo CD applications (sync and health status, deployed revision, sync history, failing resources, events, conditions), Flux kustomizations and Helm releases, Helm release history, and Kubernetes rollouts, pods, restarts and events through guarded read-only kubectl, helm and flux commands. Use this whenever the user asks whether a commit, image tag or fix is deployed to an environment, when a deploy happened or finished, why an app is OutOfSync, Degraded, Progressing or crash-looping, what changed in the last release, or whether a rollback happened, even if they do not say Argo CD, Flux, Helm or Kubernetes. Read-only: it never syncs, rolls back, restarts, scales or edits anything; it tells the user how. For build and test results use observe-lens-pipelines; for logs and metrics use observe-lens-telemetry.

Query databases read-only to see what state data is in while debugging, testing or answering a question: PostgreSQL, MySQL or MariaDB and MongoDB (including AWS RDS, Aurora and DocumentDB), directly or through an SSH tunnel, with a guard that only lets SELECT-style SQL and read-only Mongo expressions through. Discovers tables, columns and collections before querying. Use this whenever the user gives an order, customer, account, request or record id and wants to know what was stored, which status or error a record has, whether a migration or job changed data, or why the data does not match what the UI or an API shows, even if they do not say database, SQL, psql or Mongo. Read-only: for writing migrations, changesets, DDL or data fixes use a development skill instead.

Read CI pipelines read-only to explain what a build, test run or workflow did and why: Jenkins jobs and builds, GitHub Actions workflows, runs, jobs and steps, their logs, parameters, artifacts and JUnit or Allure test reports, compared run to run. Use this whenever the user pastes a Jenkins or GitHub Actions URL or a run or build number, asks why a pipeline, build, test, sanity or regression run failed or is flaky, which commit or image a run produced, whether a merge or fix passed CI, or what changed between two runs, even if they do not say Jenkins, GitHub Actions or CI. Read-only: it never triggers, reruns, cancels or approves anything; it tells the user how. Not for writing or editing pipeline files, and for what is deployed where use observe-lens-cd.

Query observability tools read-only for runtime evidence: logs (Loki, LogQL), traces (Tempo, TraceQL, spans and timings), metrics (Prometheus, PromQL: error rates, latency, restarts, CPU and memory), firing alerts and dashboards, through Grafana or directly against Prometheus, Loki and Tempo. Use this whenever the user asks what a service logged, why a request, order, call or job failed at runtime, which downstream call was slow or errored, whether errors or latency spiked, whether pods restarted, what an alert means, how a release changed error rates, or wants a dashboard or Explore link, even if they only give a trace id, request id, error message or time window and do not say Grafana, Loki, Tempo or Prometheus. Read-only; not for building or editing dashboards, alert rules or silences.