Agent attribution
also: cross-agent attribution · AI agent accountability
Agent attribution is the assignment of code changes to the specific AI agent that produced them, across all the coding tools a team uses. It answers "which agent wrote this?" in a vendor-neutral way, so multi-tool teams get one consistent accountability record instead of fragmented, per-vendor telemetry.
The multi-agent reality
A modern team rarely standardizes on one AI tool. One engineer drives Claude Code, another lives in Cursor, a third leans on Copilot, and CI runs Codex. Each vendor offers — at best — visibility into its own usage. Stitching those siloed views together by hand is impossible, and the gaps are where accountability is lost.
Agent attribution solves this by capturing every agent into one record with a shared schema. "Which agent wrote this line" gets the same answer whether the agent was Claude Code or Windsurf, and the record is comparable across tools.
Why a vendor-neutral layer is required
- Vendor telemetry is partial — for example, GitHub Copilot's audit log explicitly excludes local client session data such as prompts.
- Schemas differ per vendor, so usage data cannot be compared or aggregated reliably.
- Vendor data lives in vendor systems, creating lock-in and an external dependency for your audit trail.
How AgentDiff does cross-agent attribution
AgentDiff hooks into each agent's native callback — Claude Code's PostToolUse, Cursor's afterFileEdit, Copilot's extension, Codex's notify hook, and others — and normalizes every event into one attribution record stored in git. Adding a new agent is a small, open-source integration, so coverage grows with the ecosystem rather than depending on any single vendor.
Frequently asked questions
What does agent attribution mean?+
It means assigning each code change to the specific AI agent that produced it, consistently across every tool a team uses, rather than relying on each vendor's separate and partial telemetry.
How is agent attribution different from AI code attribution?+
They overlap heavily. AI code attribution is the broad practice of recording that code was AI-authored; agent attribution emphasizes the cross-agent, vendor-neutral dimension — getting one comparable record across many different agents.
Which AI agents can be attributed?+
AgentDiff supports Claude Code, Cursor, GitHub Copilot, Windsurf, OpenCode, Codex CLI, and Gemini/Antigravity today, and new agents can be added with a small open-source integration.
See line-level provenance on a real repo.
AgentDiff records which agent wrote which line, signs it, and keeps it in your git history. Open the live dashboard or book a walkthrough.