Skip to main content

AgentXchain v2.109.0

v2.109.0 ships governed gate actions as a complete operator surface — from runtime execution through dashboard failure detail — and hardens the release pipeline for rerun safety.

Released: 2026-04-16

What Changed​

  • Governed gate actions — gates can now own post-approval automation commands (gate_actions on gates.<gate_id>). Actions execute in order after human approval, with per-action exit-code tracking, structured failure evidence in the decision ledger, and automatic run-blocking on failure.
  • Gate-action dry-run — agentxchain approve-transition --dry-run and agentxchain approve-completion --dry-run now preview configured gate actions without executing them, so operators can verify automation before committing to approval.
  • Gate-actions operator guide — new /docs/gate-actions page covering config syntax, execution semantics, dry-run, failure/retry model, environment variables, decision-ledger evidence, and a worked example with repo-owned wrapper scripts.
  • Dashboard gate-action visibility — the repo-local dashboard now surfaces configured gate actions on pending gates and full failure detail on blocked states, including per-action status, exit codes, stderr tail, and gate-type-correct recovery hints (approve-transition --dry-run vs approve-completion --dry-run).
  • Dashboard gate-action E2E proof — real CLI-produced gate-action failures verified through the dashboard bridge and render pipeline, not just mocked fixtures.
  • Release idempotency audit — comprehensive audit of all 16 release/publish/deploy scripts and workflows. The only non-idempotent workflow (publish-vscode-on-tag.yml) now detects existing Marketplace publications and skips, matching the npm workflow pattern.
  • Approval SLA reminder notifications — stale pending approvals now emit reminder notifications so webhook-driven operators get paged when gates are waiting too long.
  • Timeout pressure during approval waits — operators now see explicit timeout countdown context while waiting for approvals.
  • Conflict-loop recovery UX parity — conflict-loop recovery guidance is now consistent across all operator surfaces.

Why This Matters​

Gate actions turn human-approved gates from a "pause and wait" checkpoint into a "pause, approve, then automate" checkpoint. This is the primitive that lets operators attach release scripts, deployment commands, notification dispatches, or any other post-approval automation to governed gates — without leaving the governed protocol.

The dashboard was the last blind spot. An operator who approves a gate from the dashboard and watches the action fail now sees exactly which action failed, why, the stderr output, and a gate-type-correct dry-run command to debug it. That is the difference between "something broke" and "action 2 of 3 failed with exit code 1, here is stderr, run this to debug."

Proof​

  • .planning/GATE_ACTIONS_SPEC.md
  • .planning/GATE_ACTIONS_DOCS_SPEC.md
  • .planning/DASHBOARD_GATE_ACTIONS_SURFACE_SPEC.md
  • cli/test/gate-actions.test.js
  • cli/test/gate-evaluator.test.js
  • cli/test/governed-ide-approval.test.js
  • cli/test/run-completion.test.js
  • cli/test/dashboard-gate-actions.test.js
  • cli/test/e2e-dashboard-gate-actions.test.js
  • cli/test/docs-gate-actions-content.test.js

Evidence​

  • 5166 tests / 1073 suites / 0 failures
  • cd website-v2 && npm run build