Agent Recovery 0.2 — Deterministic Recovery State Machines

Agent Recovery 0.2 moves the normative recovery contract from Markdown frontmatter to RECOVERY.yaml: a portable, deterministic statechart interpreted by a trusted recovery controller.

Core thesis: Let AI reason during normal operation. When failure or uncertainty occurs, switch to deterministic recovery mode.

This RFC adds:

• states, events, registered guards, actions, and deterministic transitions
• a restricted statechart profile inspired by SCXML, without embedded scripts or expressions
• an explicit Action Ledger and controller boundary
• semantic validation, including rejection of retries from unknown state
• procurement, email-delivery, and production-deployment recovery machines
• updated architecture, specification, conformance model, and project artwork

Autonomous execution. Deterministic recovery. A response is not proof of an outcome.

Agent Recovery 0.1 RFC

First public Request for Comments for RECOVERY.md, a portable recovery-plan format for side-effecting AI-agent operations.

Core rule: when an operation may have changed external state, verify the authoritative state before retrying.

This release includes the draft specification, JSON Schema, conformance levels, automated validation, and procurement, email-delivery, and production-deployment examples.

We welcome real workflow tests, failure cases, interoperability feedback, and implementation experiments.