Skip to main content

AgentXchain v2.47.0

v2.47.0 ships the run provenance surface and corrects terminal-state bootstrap behavior for continuation and recovery runs.

Run Provenance

Governed runs now carry explicit provenance metadata: trigger, parent_run_id, trigger_reason, and created_by. This closes the observability gap for long-horizon delivery chains where operators need to trace why a run was started and what it continues or recovers from.

CLI flags

FlagDescription
--continue-from <run_id>Start a fresh run from a prior completed run, with trigger=continuation
--recover-from <run_id>Start a fresh run from a prior blocked run, with trigger=recovery

Lineage query

agentxchain history --lineage <run_id> walks the parent_run_id chain to show the full ancestry of a run.

Provenance in operator surfaces

  • agentxchain status now shows an Origin: line for non-manual provenance (e.g., continuation from run_abc ("Continue after blocked review"))
  • agentxchain status --json includes a top-level provenance field
  • agentxchain report renders provenance in text, markdown, and JSON formats
  • agentxchain export includes normalized provenance in summary.provenance
  • agentxchain history shows a Trigger column with manual, continuation, recovery, intake, schedule, coordinator, or legacy

Terminal-State Bootstrap Fix

agentxchain run --continue-from and --recover-from now correctly bootstrap a fresh run envelope from terminal state. Previously, provenance-linked runs from a completed or blocked run reused the terminal state instead of initializing a new run, resulting in zero turns executed and no provenance written.

Behavior

  • agentxchain run from completed state starts a fresh manual run
  • agentxchain run --continue-from <run_id> from any terminal state starts a fresh continuation run
  • agentxchain run --recover-from <run_id> from blocked state starts a fresh recovery run
  • Fresh bootstrap resets phase, gate status, budget, and active turns from config truth

CLI Truth Corrections

  • --recover-from help text no longer references failed state (only blocked is a valid recovery source)
  • history --status filter help no longer lists failed as a filterable status

Evidence

  • 3462 tests / 741 suites / 0 failures
  • Provenance E2E: 3 subprocess tests proving continuation, recovery, and plain rerun from terminal state
  • Docusaurus production build clean