[gkarthi-signoz]

Adds a dashboard template for Cline, the open-source AI coding agent for VS Code (67k stars, 5.1M marketplace installs).

Where the telemetry comes from

Cline ships an OpenTelemetry exporter in the VS Code extension. It emits logs/events and metrics only, no traces, all under service.name cline.

Two things about it are load-bearing for this dashboard and are not in Cline's own docs:

• CLINE_BUNDLE_OVERRIDE=legacy is required. v4.1.16 ships two internal bundles. On the default next bundle, task execution runs through an SDK layer whose telemetry never reaches the extension's OTel provider, so no task.* events and no cline.* metrics are emitted at all. Verified by A/B: five varied prompts on next produced 5 event types and zero metric datapoints, while a single prompt on legacy produced 20 event types and the full metric set.
• There is an environment-variable configuration path. Cline's docs describe OTel as enterprise-only, configured through Remote Configuration in its dashboard. TelemetryProviderFactory.getDefaultConfigs() also reads CLINE_OTEL_* variables unconditionally, so this works on a stock marketplace install with no Cline account.

Event and metric shape

Panels read the log body as the event name, since Cline puts it there rather than in an attribute. ulid is the task correlation key; there is no trace or span id.

Event	Key attributes
task.tokens	tokensIn, tokensOut, cacheReadTokens, cacheWriteTokens, totalCost, model, provider
task.tool_used	tool, success, autoApproved, modelId
task.conversation_turn	mode, model, source
task.provider_api_error	errorType, errorClass, errorMessage, failurePhase

Metrics used: cline.api.duration.seconds and cline.api.ttft.seconds.

Panels

Section	Panels
Overview	Input, output and cache-read tokens; total cost; tasks, tool calls, turns, provider API errors
Tokens & Cost	Token usage over time, input tokens by model, cost by model, cache read vs write
Latency	API duration p50/p95/p99, time to first token p95 by model
Agent Activity & Errors	Event mix, tool usage by tool and approval, recent provider API errors

A Model variable scopes the token, cost and latency panels.

Query scoping worth knowing

• Token and cost panels aggregate the numeric attributes on task.tokens directly rather than reading the cline.tokens.* counters. Cline's meter only exports an instrument after its first recording, so the first datapoint already holds the post-call value and increase over a short session reads 0. The event attributes are exact from the first task.
• Latency has no event-based equivalent, so those two panels use the histograms and do need sustained traffic before percentiles mean anything.
• task.tool_used carries the model under modelId, not model. A filter written against the token panels will silently drop tool events.
• totalCost is computed client side by Cline from its own price table. It reads 0 on Cline's bundled provider and populates with a directly priced model.

Docs link

The README links https://signoz.io/docs/cline-monitoring/, which does not exist yet. The guide is being written and will use that slug. Note that a Cline guide is distinct from anything covering the Cline API.

[github-actions]

📊 Dashboard files added in this PR

• cline/cline-dashboard.json

✅ An issue has been created in SigNoz/signoz.io to track these dashboard changes.