Skip to main content

AgentXchain v2.19.0

2.19.0 fixes the first 60 seconds of evaluation.

Before this release, the binary had already grown credible governed-execution depth, but the front door still failed the evaluator:

  • the fastest path to value was hidden behind docs and scaffold steps
  • the shipped demo narrative looked like a toy counter app, which undersold why governance matters
  • baseline evidence files like AGENT-TALK.md and derived review/report artifacts could dirty the repo and block the next actor for the wrong reason

Upgrade:

npm install -g [email protected]

What changed since 2.18.0​

agentxchain demo is now the real first-run path​

The CLI now ships agentxchain demo, a one-command governed walkthrough that:

  • creates a temporary git workspace
  • runs PM → Dev → QA through the real runner interface
  • stages real turn results, decisions, objections, and phase-gate artifacts
  • cleans up the workspace automatically on success or failure

This is not a fake “hello world” transcript. It is a fast, bounded first-run experience that shows what governed multi-agent delivery actually looks like before asking evaluators to scaffold their own repo.

The adoption funnel is demo-first across the front door​

npx agentxchain demo now leads the public entry surfaces:

  • root README.md
  • cli/README.md
  • homepage hero CTA
  • quickstart Path 0

The init path still exists, but it is now the second step after the evaluator has seen governance in action. That is the right order. Asking new users to bootstrap a repo before they understand the product was avoidable friction.

The demo now demonstrates governance pressure, not coding-101​

The demo narrative moved from a toy counter app to an auth token rotation service with:

  • rollback safety
  • expiry/rotation correctness
  • audit logging

That matters because the objections now carry real delivery consequences:

  • PM catches a missing rollback plan that could invalidate live API keys
  • Dev catches a clock-skew risk the PM missed
  • QA catches a compliance/audit gap neither of the earlier roles raised

The summary now lands the actual value proposition: governed challenge caught three issues that otherwise would have shipped silently.

Baseline evidence paths no longer create false dirty-state failures​

Governed delivery depends on artifact evidence, but evidence should not sabotage the next turn.

2.19.0 hardens that contract:

  • .planning/AGENT-TALK.md, .agentxchain/reviews/, and .agentxchain/reports/ are treated as baseline-exempt evidence paths, not invisible operational paths
  • unchanged pre-existing evidence dirt is filtered from both same-HEAD and head-changed observation
  • authoritative follow-up turns remain accountable for their own changes without being blocked by stale derived evidence

This closes a real workflow integrity issue instead of just polishing output.

Why this release matters​

2.18.0 strengthened execution proof. 2.19.0 makes that proof discoverable.

That is a real product boundary:

  • evaluators can now see governance in one command
  • the demo now shows why governed challenge matters
  • baseline evidence handling no longer punishes routine collaboration artifacts

AgentXchain cannot claim adoption readiness if the product value is buried under setup work or if its own collaboration files break the next governed turn.

Where to go next​

Evidence​

  • 2315 node tests / 499 suites, 0 failures.
  • 758 Vitest tests / 36 files, 0 failures.
  • Docusaurus production build passes.
  • agentxchain demo completes in ~1.1s with 5 decisions and 3 objections.