Skip to main content

AgentXchain v2.68.0

v2.68.0 closes a real observability gap in governed rejection handling. Operators no longer need to cross-reference .agentxchain/state.json just to answer why a turn was rejected.

What Changed​

  • turn_rejected events now carry the missing audit detail: rejection payloads include reason and failed_stage, plus validation_errors when those errors actually exist.
  • The human-readable events command now exposes that detail inline: rejection rows render as reason (failed_stage) so the operator-facing surface is useful even without --json.
  • The operator surface is now frozen by proof and docs: CLI regression coverage asserts the rendered rejection detail, and the CLI reference plus standalone spec now describe the shipped behavior directly.

Why It Matters​

Claude's underlying payload enrichment was the right slice, but it was still one refactor away from disappearing in practice because the text-mode events surface had no direct regression test and no public docs note. That is not good enough for an observability claim.

v2.68.0 makes the audit trail real at both layers: the JSON event payload now contains the rejection context, and the default human-readable event stream surfaces it immediately.

Evidence​

  • 3699 tests / 796 suites / 0 failures (cd cli && npm test)
  • 82 targeted tests / 13 suites / 0 failures (node --test cli/test/governed-cli.test.js cli/test/rejection-event-payload.test.js cli/test/docs-cli-governance-content.test.js)
  • cd website-v2 && npm run build -> clean production build

Install / Upgrade​

npm install -g [email protected]
# or
brew upgrade agentxchain