TRACE FORMAT · Profile revision 1
Cursor Agent Trace
A vendor-neutral JSON record for attributing AI-generated code alongside human authorship in version-controlled files.
Why it is in the map
It contributes concrete file-modification and code-provenance semantics absent from general span formats, while remaining clearly narrower than an execution trace.
Tracevity boundary
A documented field can support reconstruction, but its presence alone does not prove completeness, authenticity, authorization, or external settlement.
Profile findings
What the documentation can—and cannot—establish
Each row distinguishes source evidence from Tracevity's bounded interpretation. “Not documented” describes the reviewed sources; it does not prove an implementation cannot record the artifact.
System identity
Partially documentedTool/model attribution and VCS-linked record identity are documented.
Tracevity interpretation
This supports code-attribution context, not a complete agent instance, definition, session, or subagent identity model.
Limit
Product/runtime identity and stable agent relationships are incomplete.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code Version 0.1.0 record schema and examples
Principal and delegation
Partially documentedHigh-level authorship category is documented.
Tracevity interpretation
Authorship category is not authenticated identity or delegated authority.
Limit
Identity-provider and authorization evidence require separate records.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code Contributor type definitions
Instruction and context
Not documentedInstruction and context contents are not standardized.
Tracevity interpretation
A link is not proof that referenced context remains available, versioned, complete, or unchanged.
Limit
Context reconstruction depends on external referenced systems.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code Conversation and related-resource link fields
Decision artifacts
Not documentedDecision artifacts are outside the format.
Tracevity interpretation
None may be inferred from attributed file ranges.
Limit
The format is intentionally focused on code attribution rather than execution reasoning.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code Record schema and non-goals
Model activity
Partially documentedA model identifier is representable.
Tracevity interpretation
Model attribution is partial and does not reconstruct inference activity.
Limit
No model request/response sequence or usage evidence is provided.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code Contributor model_id field
Tool and MCP activity
Not documentedTool execution activity is not standardized beyond contributor-tool attribution.
Tracevity interpretation
Tool provenance must not be expanded into execution reconstruction.
Limit
MCP and arbitrary external tool activity remain outside the format.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code Tool identity and record schema
Effects
Partially documentedA specialized file-modification attribution artifact is documented.
Tracevity interpretation
This can support code-change attribution but does not prove runtime execution, deployment, or external settlement.
Limit
Other effect types and independently verified resulting state are outside the RFC.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code Files, line ranges, VCS revision, and content_hash fields
Human control
Partially documentedContribution composition can identify human involvement at a high level.
Tracevity interpretation
Mixed or human authorship does not establish authorization, approval timing, or intervention.
Limit
Operational human-control events require another trace source.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code Contributor type definitions
Outcome
Partially documentedA code-attribution outcome is representable.
Tracevity interpretation
Attributed code does not establish that a requested action succeeded, ran, deployed, or changed another system.
Limit
Partial, failed, ambiguous, rollback, and compensation outcomes are outside the format.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code File attribution ranges and VCS context
Trace integrity
Partially documentedLimited content/revision linkage is documented.
Tracevity interpretation
An optional hash or VCS reference is not full-record signing, immutable custody, or independent verification.
Limit
Clock provenance, hash requirements, storage mutability, and origin authenticity remain unspecified.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code Record timestamp, VCS revision, content_hash, and storage guidance
Portability
DocumentedA versioned, machine-readable JSON interchange format is documented.
Tracevity interpretation
Syntax is portable, while mapping to execution traces and preservation through history rewrites remain unresolved.
Limit
Storage, import destinations, conversions, semantic loss, merge, and rebase behavior are not standardized.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code JSON Schema, format version, MIME type, and vendor metadata
Privacy
Not documentedPrivacy lifecycle controls are not documented by the format.
Tracevity interpretation
Storage is implementation-defined, so sensitive-data handling and custody must remain unknown.
Limit
Prompt/output/tool retention, secrets, PII, code retention, screenshots, and deletion require implementation-specific evidence.
Primary evidence (1)
- Cursor: Agent Trace - A standard format for tracing AI-generated code Record fields and storage guidance
Reconstruction reading
Do not collapse these findings into one score.
This profile describes documented evidence surfaces. Whether they are sufficient depends on the reconstruction question and the other identity, authorization, tool, and destination records available.
Potentially useful evidence
- Identity: Tool/model attribution and VCS-linked record identity are documented.
- Principal: High-level authorship category is documented.
- Models: A model identifier is representable.
- Effects: A specialized file-modification attribution artifact is documented.
- Human control: Contribution composition can identify human involvement at a high level.
Explicit gaps or uncertainty
- Context: Context reconstruction depends on external referenced systems.
- Decisions: The format is intentionally focused on code attribution rather than execution reasoning.
- Tools: MCP and arbitrary external tool activity remain outside the format.
- Privacy: Prompt/output/tool retention, secrets, PII, code retention, screenshots, and deletion require implementation-specific evidence.
Source register
1 reviewed primary sources
- Agent Trace - A standard format for tracing AI-generated codeCursor · OFFICIAL REPOSITORY · observed August 28, 2026 · currentOpen source ↗
Next question
What evidence does your use case require?
Move from documented field presence to an explicit reconstruction target, or examine selected directional format mappings without changing this system's documentation posture.
Trace Reconstruction Requirements Explore compatibility evidence