Problem
Copilot CLI emits useful native OpenTelemetry spans for agent activity, model calls, and tool execution. The observed spans already include gen_ai.* and github.copilot.* data such as conversation IDs, model selection, tool names, and input/output/cache/reasoning token usage.
However, the agent/model cannot enrich those native spans with delivery context. Span creation and export belong to the Copilot CLI runtime, and Copilot CLI currently exposes no supported agent/runtime API or hook to set attributes on its native OTEL spans. Custom instructions, custom agents, MCP servers, and tools cannot reliably fill this gap.
Requested capability
Provide a governed runtime API/hook that allows a Copilot CLI agent (or approved integration) to add allow-listed custom attributes to native Copilot CLI OTEL spans before export.
Illustrative API:
on_span_created(span)
span.set_attribute(key, value)
Goal and use cases
Enable end-to-end correlation of AI-assisted engineering activity with work items, Git changes, and delivery outcomes. An enriched trace/span should support data points such as:
workitem.id
vcs.repository.url.full
vcs.repository.ref.name
vcs.repository.ref.revision
gitlab.merge_request.id
ai.session.id
gen_ai.conversation.id
gen_ai.request.model
gen_ai.response.model
gen_ai.usage.input_tokens
gen_ai.usage.output_tokens
gen_ai.usage.cache_read.input_tokens
gen_ai.usage.cache_write.input_tokens
gen_ai.usage.reasoning.output_tokens
gen_ai.tool.name
The primary correlation key is repository identity plus commit SHA. That joins Copilot traces to Git commits, merge requests, pipelines, artifacts, deployments, and production outcomes. Ticket and session identifiers provide supporting context. The capability should also support integrations with ticketing systems and approved software-supply-chain/provenance standards.
Requirements
- Attribute mutations occur in the CLI runtime before export, not through model prompting alone.
- Allow-list attribute keys and constrain types, value sizes, and cardinality.
- Support privacy/redaction controls and organization policy.
- Make provenance clear: identify whether an attribute was supplied by the runtime, agent, or integration.
- Apply configured context consistently to relevant native spans, including
invoke_agent, chat, and execute_tool.
- Preserve existing OTEL semantic conventions where applicable, especially
vcs.* and gen_ai.*.
Workaround
An OpenTelemetry Collector or OTLP proxy can enrich exported spans externally today. This works for process-wide Git/Jira context, but does not provide agent-aware, per-span enrichment at the point where Copilot CLI creates the spans.
Observed version
GitHub Copilot CLI 1.0.88.
Problem
Copilot CLI emits useful native OpenTelemetry spans for agent activity, model calls, and tool execution. The observed spans already include
gen_ai.*andgithub.copilot.*data such as conversation IDs, model selection, tool names, and input/output/cache/reasoning token usage.However, the agent/model cannot enrich those native spans with delivery context. Span creation and export belong to the Copilot CLI runtime, and Copilot CLI currently exposes no supported agent/runtime API or hook to set attributes on its native OTEL spans. Custom instructions, custom agents, MCP servers, and tools cannot reliably fill this gap.
Requested capability
Provide a governed runtime API/hook that allows a Copilot CLI agent (or approved integration) to add allow-listed custom attributes to native Copilot CLI OTEL spans before export.
Illustrative API:
Goal and use cases
Enable end-to-end correlation of AI-assisted engineering activity with work items, Git changes, and delivery outcomes. An enriched trace/span should support data points such as:
The primary correlation key is repository identity plus commit SHA. That joins Copilot traces to Git commits, merge requests, pipelines, artifacts, deployments, and production outcomes. Ticket and session identifiers provide supporting context. The capability should also support integrations with ticketing systems and approved software-supply-chain/provenance standards.
Requirements
invoke_agent,chat, andexecute_tool.vcs.*andgen_ai.*.Workaround
An OpenTelemetry Collector or OTLP proxy can enrich exported spans externally today. This works for process-wide Git/Jira context, but does not provide agent-aware, per-span enrichment at the point where Copilot CLI creates the spans.
Observed version
GitHub Copilot CLI
1.0.88.