Plan and verify Application Insights telemetry across services, queues, privacy filters, sampling, cost controls, and rollout gates.
An SDK can be installed and still leave a production system hard to diagnose. Services may emit separate traces that never join across a queue. A query parameter can quietly copy customer data into telemetry. Two instrumentation paths can double ingestion while the dashboard still looks plausible.
Application Insights Instrumentation turns setup into a controlled production rollout. It starts with the decisions telemetry must support, maps critical journeys to traces, metrics, logs, exceptions, dependencies, and events, and assigns stable service, environment, version, route, and dependency identity. It covers synchronous and asynchronous correlation, including producer injection and consumer extraction across queues.
The workflow keeps privacy, cardinality, sampling, retention, and cost visible before rollout. Sensitive sources are blocked at collection. Missing volume inputs stay marked cost-unbounded. Every change remains proposed, executed-unverified, or verified based on actual evidence. Test, canary, production, acceptance, and rollback states are separate, so installing a package never counts as proof that production is observable.
A canonical JSON handoff contract records scope, authority, services, boundaries, signals, privacy controls, volume assumptions, changes, acceptance evidence, rollback, limitations, verdict, and one next action. Recovery guidance covers missing data, broken correlation, duplicate telemetry, wrong service identity, sensitive-field exposure, and cost spikes.
Version 1.0.0 was reviewed on 2026-07-17 and evaluated through Codex CLI with gpt-5.6-sol in 12 fresh read-only contexts. The candidate scored 100 across three cases, compared with 69.58 for the same agent without the skill and 84.58 for the exact free upstream. All cases passed. It led the upstream by 13.75 points on the API and worker rollout plan, 18.75 points on the canonical instrumentation record, and 13.75 points on unsafe-canary recovery. The evaluation covers those supplied cases. It does not prove that an Azure tenant, resource, telemetry stream, package version, privacy control, ingestion budget, or deployment matches supplied evidence.
The reviewed package SHA-256 is 5a96232fa76c5935700b5a48ce8d2806f7e1cb1165cdae68476eaf09daf26615. The package does not contact Azure, prove tenant permissions, validate live telemetry, change retention, delete stored data, approve a deployment, or replace security, privacy, legal, or incident review. Maintenance and install support are handled by the platform-operated publisher for this listing.
Input
A Node API sends checkout work through Service Bus to a worker and Stripe. The team needs end-to-end failure and latency evidence without collecting customer or payment data. Azure and deployment access are unavailable in this session.
Output
A read-only signal and boundary plan with stable service identity, W3C propagation, privacy allowlists, cardinality controls, unknown cost inputs, staged acceptance gates, rollback triggers, and no claim that telemetry is live.
Input
Return only JSON for two Node services, one queue boundary, supplied runtime and hosting details, incomplete Azure identifiers, no executed changes, and an approved list of sensitive sources to exclude.
Output
Provide the application and service inventory, runtime and hosting model, critical journeys, Azure scope if known, current instrumentation, authority, sensitive-data rules, traffic or volume evidence, budget or retention constraints, and the action you want: assess, plan, implement, verify, or recover. The skill returns an evidence-bound plan or handoff record and stops before an unapproved mutation or unsupported production claim.
Repository or supplied architecture, application version, runtime, framework, hosting model, deployment units, environments, regions, and critical user or background journeys.
Known tenant, subscription, resource group, Application Insights resource, workspace, current SDKs or agents, configuration, service identity, and observation times. Unknown identifiers may remain explicit.
Sensitive-data exclusions, allowed attributes, sampling or retention rules, traffic and payload evidence, approved budget or cap, and the owners for privacy, security, and cost decisions.
Repository, Azure read, configuration, deployment, and production-approval authority plus canary scope, acceptance window, last known safe configuration, and rollback owner.
SKILL.md; production instrumentation workflow covering telemetry coverage, privacy, correlation, sampling, cost, staged rollout, acceptance, rollback, and recovery; canonical instrumentation-plan contract; Agent Skills interface metadata
No reviews yet.
Contract-valid JSON binding scope, authority, journeys, services, boundaries, signals, privacy, cost-unbounded inputs, proposed changes, not-run acceptance checks, rollback, limitations, verdict, and one owner-bound next action.
Input
A new OpenTelemetry distro runs beside a legacy agent. Counts doubled, a sensitive query field appeared, queue traces do not correlate, and projected ingestion exceeds the cap. Only pausing or rolling back the canary is approved.
Output
A blocked full rollout, separate duplicate/privacy/correlation/cost findings, early-source containment, approved pause or rollback plan, authoritative stored-data escalation, re-verification matrix, and no claim that a mutation or deletion occurred.
Creator
Fferavo93
