Skip to main content

AgentXchain v2.154.10

v2.154.10 is the downstream full-auto recovery release after the accepted BUG-52 fix. It targets the next tusq.dev blockers: continuous mode idling instead of continuing an active governed run, and launch_ready unblock redispatching product marketing instead of completing the run.

Bug Fixes​

  • Continuous current-run continuation: run --continuous now continues the active governed run even when active_turns is empty. An active phase waiting for its next assignment is real work, not a vision idle cycle.

  • Terminal gate unblock completion: unblock <hesc> now reconciles final standing gates before dispatch. If a launch turn re-verified complete launch_ready artifacts, returned needs_human, changed no files, and routed to human for run-completion approval, AgentXchain marks the terminal gate passed and emits completion instead of assigning another launch turn.

Evidence​

  • node --test cli/test/beta-tester-scenarios/bug-59-full-auto-gate-closure.test.js cli/test/run-schedule-e2e.test.js cli/test/continuous-run.test.js cli/test/continuous-run-e2e.test.js cli/test/beta-tester-scenarios/bug-54-qa-cli-chain-reliability.test.js cli/test/beta-tester-scenarios/bug-61-tester-quoteback-ask-content.test.js -> 56 tests / 13 suites / 0 failures / 0 skipped
  • node --test cli/test/beta-tester-scenarios/bug-52-gate-unblock-phase-advance.test.js cli/test/beta-tester-scenarios/bug-53-continuous-auto-chain.test.js -> 24 tests / 2 suites / 0 failures / 0 skipped

Tester Re-Run Contract​

Run the shipped package:

npx --yes -p [email protected] -c "agentxchain --version"

For downstream closure, start from the same clean tusq.dev checkpoint used for the v2.154.9 BUG-52 acceptance and quote:

  • run --continuous --vision .planning/VISION.md --max-runs 10 --max-idle-cycles 3 --poll-seconds 5 --triage-approval auto --auto-checkpoint --no-report continues the active governed run instead of ending with runs_completed=0 and idle_cycles=3
  • the current run reaches QA and launch through the real adapter path; quote any startup_latency_ms, elapsed_since_spawn_ms, and runtimes.<id>.startup_watchdog_ms evidence relevant to BUG-54
  • approval-policy evidence is quoted when the project config contains approval_policy; if status still reports approval_policy: null, say so explicitly rather than fabricating ledger rows
  • launch_ready approval after a no-change launch re-verification marks launch_ready=passed
  • final state is status=completed, pending_run_completion=null, and active_turns={}
  • durable events include gate_approved for run_completion and run_completed
  • phase-transition recovery still quotes phase_entered with reconciled_before_dispatch when a non-terminal phase advances
  • no duplicate product_marketing dispatch occurs after terminal human approval
  • session_continuation rows are quoted if the continuous session actually chains to a next run; otherwise quote the terminal reason

For BUG-55 regression proof, continue using accept-turn followed by checkpoint-turn and quote verification.produced_files, undeclared_verification_outputs, and git status --short.

Clean tree means BUG-55 is fixed for your reproduction.

Keep BUG-62 separate in scratch because prior tusq.dev history mixed checkpoint baselines with governed .agentxchain changes.