Skip to content

zzhang82/Agent-Memory-Bridge

v0.22.2 Feature

This release adds 2 notable features for engineering teams evaluating rollout.

Published 8d MCP Developer Tools
✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agent-memory ai-agents claude cline codex coding-agents
+10 more
cursor developer-tools llm local-first mcp mcp-server model-context-protocol persistent-memory python sqlite

Summary

AI summary

Onboarding flow now uses a consistent .amb-venv baseline across Windows, macOS, and Linux.

Changes in this release

Refactor Low

Standardizes install baseline to immutable v0.22.2 source archive and .amb-venv environment name across Windows, macOS, Linux.

Standardizes install baseline to immutable v0.22.2 source archive and .amb-venv environment name across Windows, macOS, Linux.

Source: llm_adapter@2026-07-19

Confidence: high

 —
Refactor Low

`first-run` derives venv interpreter for both Windows and POSIX hosts, rendering appropriate shell commands.

`first-run` derives venv interpreter for both Windows and POSIX hosts, rendering appropriate shell commands.

Source: llm_adapter@2026-07-19

Confidence: high

 —
Refactor Low

Distinguishes placeholder-safe `--example` output from runnable config using approved local paths in guides and renderer.

Distinguishes placeholder-safe `--example` output from runnable config using approved local paths in guides and renderer.

Source: llm_adapter@2026-07-19

Confidence: high

 —
Refactor Low

Clarifies that `doctor` checks prerequisites, `verify` launches isolated stdio runtime, and MCP status confirms config loading without performing registration.

Clarifies that `doctor` checks prerequisites, `verify` launches isolated stdio runtime, and MCP status confirms config loading without performing registration.

Source: llm_adapter@2026-07-19

Confidence: high

 —
Refactor Low

First memory‑loop instructions identify `store(...)` and `recall(...)` as MCP tool calls rather than terminal commands.

First memory‑loop instructions identify `store(...)` and `recall(...)` as MCP tool calls rather than terminal commands.

Source: llm_adapter@2026-07-19

Confidence: high

 —
Refactor Low

Cross‑platform tests now use synthetic fixtures covering POSIX shell quoting and Windows PowerShell command rendering.

Cross‑platform tests now use synthetic fixtures covering POSIX shell quoting and Windows PowerShell command rendering.

Source: llm_adapter@2026-07-19

Confidence: high

 —
Refactor Low

Ensures public MCP surface remains unchanged at 10 tools; no new tools or memory features added.

Ensures public MCP surface remains unchanged at 10 tools; no new tools or memory features added.

Source: llm_adapter@2026-07-19

Confidence: high

 —
Refactor Low

`first-run` writes zero client‑config entries and zero durable memory records, preserving existing state.

`first-run` writes zero client‑config entries and zero durable memory records, preserving existing state.

Source: llm_adapter@2026-07-19

Confidence: high

 —
Full changelog

v0.22.2 - Onboarding Coherence

Agent Memory Bridge 0.22.2 makes the released setup flow match the public
onboarding guides. The pinned archive, installed first-run report,
client-config renderer, and verification sequence now use the same .amb-venv
baseline across Windows, macOS, and Linux.

0.22.2 = one released onboarding contract across the docs and CLI, with client registration kept explicit and manual.

What Changed

  • The baseline install uses the immutable v0.22.2 source archive and the
    .amb-venv environment name consistently.
  • first-run derives the venv interpreter for Windows and POSIX hosts and
    renders PowerShell-style Windows commands and sh-compatible POSIX commands.
  • The config renderer and integration guides distinguish placeholder-safe
    --example output from runnable config built with approved local paths.
  • The guides distinguish local validation from client registration:
    doctor checks prerequisites, verify launches an isolated stdio runtime,
    and the client's MCP status/tool view confirms that the config was loaded.
  • The first memory-loop instructions identify store(...) and recall(...)
    as MCP tool calls rather than terminal commands.
  • Cross-platform tests use synthetic fixtures and cover POSIX shell quoting and
    Windows PowerShell command rendering.

Why It Matters

The current onboarding guidance was added after v0.22.1, so users installing
that tag could still receive older first-run output. v0.22.2 puts the public
guides and the installed CLI on the same setup contract.

Validation

  • pytest: 395 passed
  • Release gate: all nine GitHub Actions jobs must be green across Windows,
    macOS, and Linux with Python 3.11, 3.12, and 3.13
  • public MCP surface: unchanged at 10 tools
  • client-config writes from first-run: 0
  • durable memory record writes from first-run: 0

Boundaries

0.22.2 adds no MCP tools or memory features. It does not write client config,
restart clients, or prove that a client loaded its generated config. doctor
and verify remain local checks; client MCP status/tool visibility is the
registration gate.

The Phase 1 pilot will measure independent-user completion time, setup failures,
and maintainer intervention. This release does not claim those outcomes yet.

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track zzhang82/Agent-Memory-Bridge

Get notified when new releases ship.

Sign up free

About zzhang82/Agent-Memory-Bridge

MCP-native, local-first memory for coding agents that turns coding sessions into reusable engineering memory: decisions, gotchas, and domain knowledge.

All releases →

Related context

Earlier breaking changes

  • v0.21.0 Procedures with declared domains are rejected for mismatched task domains; legacy unscoped procedures now emit a warning instead of auto‑assigning policy.

Beta — feedback welcome: [email protected]