Span attributes are metadata tags on individual spans. These are the raw key-value pairs your coding harnesses emit and Dash0 reads. Use them to filter sessions (for example, dash0.gen_ai.vcs.repository.name="my-repo") and group metrics (for example, sum by (user_name)).

For the aggregated time-series metrics Dash0 derives from these spans, see Derived Metrics.

Note

Coverage varies by harness. Where an attribute is only emitted by some harnesses, the table says so. An attribute that a harness does not report is absent rather than empty, so a breakdown built on it shows nothing for that harness's sessions.

Resource Attributes

Set once per session and carried on every span it produces.

Attribute	Description
`service.name`	Harness name (`claude-code`, `cursor`, `codex`), or the `agent_name` override.
`service.version`	Plugin version (`dev` in source runs).
`gen_ai.harness.name`	The coding harness that ran the session, for example `claude-code`, `cursor`, or `codex`. Exposed as the Harness filter on every tab.
`gen_ai.agent.name`	Agent name, or the subagent type on `invoke_agent` spans.
`gen_ai.provider.name`	Model provider, for example `anthropic` or `openai`. Resolved from the model prefix, else the harness default.
`gen_ai.conversation.id`	Session identifier that groups all spans from a single session.
`dash0.team.name`	Team the session is attributed to, for example `platform`. Only set when `team_name` is configured. Telemetry without it groups under `unknown`.
`process.working_directory`	Working directory. `~`-redacted when `omit_user_info` is enabled.

VCS Context Attributes

Dash0-specific attributes capturing version control context. The branch name is what links a session to a pull request, so a session missing dash0.gen_ai.vcs.ref.head.name cannot be counted as assisting one.

Attribute	Description
`dash0.gen_ai.vcs.repository.name`	Repository name.
`dash0.gen_ai.vcs.repository.url.full`	Full repository URL, normalized to https.
`dash0.gen_ai.vcs.owner.name`	Repository owner (organization or user).
`dash0.gen_ai.vcs.provider.name`	VCS provider: `github`, `gitlab`, `bitbucket`, or `gitea`.
`dash0.gen_ai.vcs.ref.head.name`	Branch or tag name.
`dash0.gen_ai.vcs.ref.head.type`	`branch` or `tag`.
`dash0.gen_ai.vcs.ref.head.revision`	Commit SHA of the checked-out ref.
`dash0.gen_ai.vcs.commit.sha`	Commit SHA, auto-extracted from git commit output. Survives `omit_io`.
`dash0.gen_ai.vcs.pull_request.url`	Pull request or merge request URL, auto-extracted from tool responses. Survives `omit_io`.
`dash0.gen_ai.vcs.issue.url`	Issue URL, auto-extracted from tool responses. Survives `omit_io`.

Chat Span Attributes

Emitted per turn, on spans where gen_ai.operation.name is chat or invoke_agent.

Attribute	Description
`gen_ai.operation.name`	`chat` or `invoke_agent`.
`gen_ai.request.model`	Model used, for example `claude-opus-4-8` or `gpt-5.3-codex`.
`gen_ai.request.reasoning.level`	Reasoning level requested: `low`, `medium`, `high`, or `xhigh`. This is what the Cost tab's Effort breakdown groups by. Claude Code only — no other harness reports an effort field.
`gen_ai.conversation.name`	Session title. Claude Code only, derived from the first prompt and gated by `omit_io`.
`gen_ai.usage.input_tokens`	Input tokens for a model invocation.
`gen_ai.usage.output_tokens`	Output tokens for a model invocation.
`gen_ai.usage.cache_read.input_tokens`	Input tokens served from the prompt cache.
`gen_ai.usage.cache_creation.input_tokens`	Input tokens spent populating the prompt cache.
`dash0.gen_ai.usage.cache_creation.ephemeral_5m.input_tokens`	Cache creation tokens for the 5-minute tier. Claude Code only.
`dash0.gen_ai.usage.cache_creation.ephemeral_1h.input_tokens`	Cache creation tokens for the 1-hour tier. Claude Code only.
`gen_ai.usage.reasoning.output_tokens`	Thinking tokens. A subset of `output_tokens`, not an addition, so cost is unaffected. Emitted only when greater than zero.
`gen_ai.input.messages`	User messages as JSON. Gated by `omit_io`.
`gen_ai.output.messages`	Assistant messages as JSON. Gated by `omit_io`.
`gen_ai.agent.id`	Subagent ID, on `invoke_agent` spans.
`exception.message`	Error text, on a failed session.

Billing and Plan Attributes

Attribute	Description
`dash0.gen_ai.billing_mode`	`subscription`, `api`, `metered_external`, or `unknown`. Claude Code and Codex.
`dash0.gen_ai.billing_provider`	`bedrock`, `vertex`, `foundry`, or `gateway`. Claude Code only, and only when the mode is `metered_external`.
`dash0.gen_ai.plan_type`	Provider plan name. Provider vocabulary: display it, do not parse it. Claude Code and Codex.
`dash0.gen_ai.rate_limit.{primary,secondary}.used_percent`	Rate limit consumption, 0–100. Codex only.
`dash0.gen_ai.rate_limit.{primary,secondary}.window_minutes`	Rate limit window length. Codex only.
`dash0.gen_ai.rate_limit.{primary,secondary}.resets_at`	Rate limit reset time, unix seconds. Codex only.
`dash0.gen_ai.credits.available` / `.unlimited` / `.balance`	Credit state. Codex only.

Tool Span Attributes

Emitted per tool call, on spans where gen_ai.operation.name is execute_tool.

Attribute	Description
`gen_ai.tool.name`	Tool name, for example `Bash` or `Read`. MCP tool names are stripped of their `mcp__<server>__` prefix.
`gen_ai.tool.type`	`function`.
`gen_ai.tool.call.id`	Tool-use ID.
`gen_ai.tool.call.arguments`	Tool input. Gated by `omit_io`, capped at 16 KB.
`gen_ai.tool.call.result`	Tool output. Gated by `omit_io`, capped at 16 KB.
`dash0.gen_ai.tool.mcp_server`	MCP server name, parsed from the tool name.
`dash0.gen_ai.tool.bash.command_family`	Leading binary from Bash commands, for example `git` or `npm`.
`dash0.gen_ai.tool.skill.name`	Skill name from Skill tool invocations.
`dash0.gen_ai.tool.skill.source`	`command` or `model`: whether a person or the model loaded the skill.
`dash0.gen_ai.code.lines_added`	Lines added, from the Edit/Write/MultiEdit patch. Claude Code only.
`dash0.gen_ai.code.lines_removed`	Lines removed, from the Edit/Write/MultiEdit patch. Claude Code only.

User Attributes

These follow the OpenTelemetry semantic conventions for user identity.

Attribute	Description
`user.name`	Developer identity, from `git config user.name` or the OS account. Replaced with a 16-character hash when `omit_user_info` is enabled.
`user.email`	Developer email. Read from git only, never inferred. Omitted when `omit_user_info` is enabled.
`dash0.gen_ai.user.identity.source`	`git` or `os`: which source `user.name` came from.

What Is Not a Span Attribute

Some values shown in AI SDLC Insights are computed by Dash0 rather than emitted by your harness, so there is no attribute to filter or group by:

Evaluations and activity are assessed by Dash0 over a sample of sessions. See Evaluation.
PR type, review depth, and the pipeline stages are derived from pull request lifecycle data supplied by the GitHub integration, not from harness telemetry.
Cost is computed from token counts against a Dash0-maintained pricing catalog. See Cost.

Using Attributes in PromQL

In PromQL, attribute names appear with dots replaced by underscores, for example dash0_gen_ai_vcs_repository_name, user_name, or gen_ai_request_model. Use them to filter sessions or group metrics (for example, sum by (user_name)).