v0.17.0 - Human Review Workflow

Agent Memory Bridge 0.17.0 is a narrow operator-workflow release over the v0.16
review queue.

0.17.0 = explicit human review workflow without automatic durable writeback.

What Changed

• Added agent-memory-bridge review-workflow, a CLI/report layered over
  review-queue.
• The workflow report turns queue items into decision prompts, manual steps,
  allowed outcomes, and blocked-until gates.
• Added a deterministic review-workflow benchmark snapshot at
  benchmark/latest-review-workflow-report.json.
• The release contract now checks review-workflow facts alongside the existing
  retrieval, calibration, task-memory, procedure, contention, adversarial,
  memory-evolution, and review-queue snapshots.
• The public MCP surface remains unchanged at 10 tools.

Validation

• pytest: 306 passed
• release contract: passed
• public MCP surface: unchanged at 10 tools
• review workflow benchmark:
  • review_workflow_source_queue_item_count = 6
  • review_workflow_item_count = 6
  • review_workflow_manual_step_count = 27
  • review_workflow_requires_human_count = 6
  • review_workflow_auto_write_count = 0
  • review_workflow_no_auto_writeback = true
  • review_workflow_public_mcp_surface_change = false
  • review_workflow_item_type_count = 6

Boundaries

This is still not:

• a review UI
• an autonomous reviewer
• an automatic promotion, deletion, merge, or writeback path
• a scheduler or worker runtime
• a new MCP tool surface

review-workflow is a proposal-only operator report. Humans or external
harnesses can use it to decide what to do, but AMB core does not execute the
durable mutation for them.

Public MCP Surface

The public MCP surface remains unchanged:

• store
• recall
• browse
• stats
• forget
• promote
• export
• claim_signal
• extend_signal_lease
• ack_signal