AgentXchain v2.15.0
2.15.0 closes the intake-to-coordinator workflow gap. Handoff, blocked-state recovery, hook-stop asymmetry, and single-repo automation continuity all ship with real subprocess E2E proofs — no forged state, no manual staging.
Upgrade:
npm install -g [email protected]
What changed since 2.14.0
Intake-to-coordinator handoff is now a proven product surface
intake handoff bridges source-repo intent to a coordinator workstream. The handoff is bound by super_run_id, renders coordinator context into artifacts, and is discoverable from all front-door surfaces.
The E2E proof drives real CLI dispatch through multi step, accept-turn, and multi approve-gate — no barrier mutations or state forgery.
Blocked coordinators can now recover
Previously, a coordinator that entered blocked state (e.g., from a post-action hook violation) had no shipped recovery path. 2.15.0 adds multi resume:
- Resyncs child repos first
- Fails closed on blocked children
- Restores
activeorpausedstate - Records
blocked_resolvedhistory entries
intake resolve now accepts blocked as a valid source state, so the same run/workstream can recover to completed.
Hook-stop asymmetry is documented and guarded
Pre-action hooks are idempotent barriers: they reject the operation but do not persist blocked state. Post-action hooks can persist blocked and fire on_escalation. This distinction was always intentional but never explicit. It is now spec'd, documented, and guarded.
Single-repo intake flows into run through the same run_id
intake start hands off to agentxchain run, which adopts the intake-started run rather than silently creating a new one. The E2E proof drives the full record -> triage -> approve -> plan -> start -> run -> resolve sequence and verifies run_id continuity.
Interface alignment barriers
Real interface_alignment barriers now ship with an end-to-end multi-repo docs example. Runner adoption docs are tightened with Tier 3 conformance requirements.
Why this release matters
This release proves the lights-out automation story end to end. A governed intent can enter the system, flow through intake, hand off to a coordinator or local runner, recover from blocked state, and resolve — all through shipped CLI surfaces with real subprocess E2E proofs.
Where to go next
- See Continuous Delivery Intake for the intake lifecycle.
- See Multi-Repo Coordination for coordinator handoff and recovery.
- See Protocol Reference for the full governed workflow contract.
Evidence
2048node tests /457suites,0failures694Vitest tests /36files,0failures- Tier 1:
46fixtures. Total conformance corpus:74fixtures - Docusaurus production build passes