ASR is a Draft, as formatVersion 0.1: complete enough to implement and test against, and open to change in response to review. Version 0.1 is fitted first to four tools: Claude Code, T3 Code, Codex and OpenClaw. Everything they record has a place, either as a standard field or record type or as a documented vendor key in the agent registry. There is no experimental tier: every feature in the specification is normal, and features that are not ready are listed below as future work.

​
Before Preview

The draft is open for review now. This work comes before the format is declared Preview:
  1. Normative algorithms, each with fixtures: the rule that tells mechanical siblings from real branches in harnesses that store sessions as trees, the active path, and the context view.
  2. A golden corpus of sanitized sessions from at least ten harnesses, plus adversarial fixtures for:
    • the active path and context view;
    • revision chains;
    • terminals;
    • declarations;
    • copies;
    • every compaction kind;
    • torn tails (incomplete final lines);
    • unknown extensions.
  3. The validator, the JSON Schemas and the agent identity registry. Draft versions of the validator, the schemas and the conformance fixtures exist; the schemas are published.
  4. The Agent Client Protocol (ACP) v2 recorder support matrix, field by field, in the recorder mapping.
  5. Two independent converters that support the same five harnesses and produce matching results on the corpus.

​
1.0

Version 1.0 requires:
  • the golden corpus and adversarial fixtures, with every example in the specification validated in CI;
  • two independent implementations.
After 1.0, changes follow the versioning rules in Governance, and the project intends to propose ASR to the Agentic AI Foundation.

​
Not in 0.1 (future work)

These are left out of the 0.1 schema:
  • principal, tenant and on-behalf-of identity;
  • payments;
  • a person taking over the agent’s browser;
  • tamper-evidence seals;
  • guardrail verdicts;
  • export coverage, retention policies and legal holds;
  • provenance verified by the provider, as distinct from content the client asserts;
  • agent-to-agent tasks across vendors (A2A);
  • live voice sessions;
  • streams of a memory store on its own;
  • records of resource access on their own;
  • dirtyPatch, a patch of the uncommitted changes in a workspace root;
  • general effect and task records.
Each can join later under the same rule as every other feature: a field or record type is standard when at least two of Claude Code, T3 Code, Codex and OpenClaw record it. See Governance.