AgentXchain v2.82.0
v2.82.0 fixes a critical approval-gate bypass, hardens the intake pipeline, rejects scaffold placeholders in gate evaluation, sweeps all 20 integration guides for truth-boundary accuracy, and expands E2E subprocess proof for release gates and coordinator blockers.
What Changed
Approval-Gate Bypass Fix (Security)
Blocked runs with pending requires_human_approval objects were incorrectly accepted by resume, step, and reactivate. All three paths now reject runs that have outstanding approval requirements. This prevents governed runs from advancing past human-required gates without explicit sign-off.
Intake Pipeline Hardening
intake startnow unconditionally rejects paused runs — previously, paused state could drift into the intake pipeline- Intake context is now preserved across
intake startinvocations, preventing context loss during signal-to-run transitions
Workflow-Kit Placeholder Rejection
Gate evaluation (section_check and system_spec gates) now detects and rejects scaffold placeholder text. Mock agents in E2E tests write real content instead of TODO stubs. Dispatch templates use schema-guided angle-bracket format (<description>) instead of ambiguous TODO placeholders.
Integration Guide Truth-Boundary Sweep
All 20 integration guides audited and corrected:
- Cursor, Windsurf: removed claims about native connectors that don't exist — both use
local_clionly - Google Jules: corrected truth boundary for API access patterns
- Amazon Bedrock: fixed factual accuracy on model identifiers and bootstrap paths
- API proxy guides: added governed bootstrap paths showing real
agentxchain init→agentxchain runflow - local_cli guides: added bootstrap paths and guard tests for setup accuracy
E2E Proof Expansion
- Release notes gate (
e2e-release-gate.test.js): 4 subprocess tests provinggate show --evaluatecatches placeholder release notes, lifecycle blocks on semantic failure, and advances after fix - Coordinator gate blockers (
e2e-coordinator-gate-blockers.test.js): subprocess proof formulti stepsurfacingrepo_active_turns,barrier_unsatisfied, andrepo_run_id_mismatchwith structured error output - Coordinator recovery:
multi stepnow prints structured[repo_run_id_mismatch]with expected/actual run IDs instead of a generic one-line error
Other Improvements
- Guided governed init prompts for better first-run experience
- Front-door scheduling discoverability in README and
llms.txt - Governed inspection commands added to both README surfaces
- Fixed 33 test failures from
import.meta.urlvsprocess.cwd()repo root resolution - Homebrew sync PR supersession drift fix
Upgrade
npm install -g agentxchain@latest
# or
brew upgrade agentxchain
Evidence
- 4233 tests / 897 suites / 0 failures (
cd cli && npm test) - All E2E subprocess tests green
- Docusaurus build clean