This release adds 1 notable feature for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Topics
+10 more
Summary
AI summaryAdded activation-receipt CLI/report for cross‑client shared memory provenance.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Security | Medium |
Receipt hashes metadata but omits raw memory content, private paths, session IDs, workspace values, and model IDs. Receipt hashes metadata but omits raw memory content, private paths, session IDs, workspace values, and model IDs. Source: granite4.1:30b@2026-07-18-audit Confidence: low |
— |
| Feature | Medium |
Adds "activation-receipt" CLI/report for cross-client activation provenance. Adds "activation-receipt" CLI/report for cross-client activation provenance. Source: llm_adapter@2026-07-18 Confidence: high |
— |
| Feature | Low |
Requires one writer memory with specific tags and one reader signal with specific tags for receipt generation. Requires one writer memory with specific tags and one reader signal with specific tags for receipt generation. Source: granite4.1:30b@2026-07-18-audit Confidence: low |
— |
| Feature | Low |
Reader signal must be acked and record the writer memory ID as observed_memory_id. Reader signal must be acked and record the writer memory ID as observed_memory_id. Source: granite4.1:30b@2026-07-18-audit Confidence: low |
— |
| Feature | Low |
Writer and reader must have distinct declared source_client labels. Writer and reader must have distinct declared source_client labels. Source: granite4.1:30b@2026-07-18-audit Confidence: low |
— |
Full changelog
v0.22.0 - Cross-Client Activation Receipt
Agent Memory Bridge 0.22.0 adds a local receipt for one shared-memory loop
across two configured coding-agent clients.
0.22.0 = client A writes reviewed project memory, client B reads it and acks a signal, and AMB emits a sanitized declared-provenance receipt.
What Changed
- Added the
activation-receiptCLI/report:
agent-memory-bridge activation-receipt --namespace project:demo --correlation-id activation-demo-001 --format markdown
- The receipt reads existing AMB rows for one namespace and correlation id. It
requires exactly one writer memory taggedworkflow:cross-client-activation,
activation-role:writer, andreviewed:true, plus exactly one reader signal tagged
workflow:cross-client-activationandactivation-role:reader. - The reader signal must be acked and must record the writer memory id as
observed_memory_id. - The writer and reader must have present, distinct declared
source_client
labels. - The receipt hashes namespace, correlation id, record ids, and source-client
labels. It does not print raw memory content, private paths, session ids,
client workspace values, or model ids.
Why It Matters
Native per-client memory is now table stakes. AMB's useful wedge is shared,
governed, inspectable continuity across coding agents. This release gives
operators a small local receipt that shows two distinct declared client labels
participated in one reviewed memory loop without adding another MCP tool.
Validation
pytest:388 passed- cross-client activation receipt tests: passed
- public MCP surface: unchanged at
10tools - durable writeback from receipt generation:
0 - client config writes from receipt generation:
0
Boundaries
This is declared provenance only. It does not authenticate client identity,
certify any vendor, prove marketplace distribution, or claim external adoption.
It does not add MCP tools, write client configuration, promote memory, or perform
automatic durable writeback.
The fixed governed-change proof remains the v0.21 proof: the report
target_release is 0.21.0, and the manifest has current_release = 0.20.0
and target_release = 0.21.0.
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
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.
Related context
Related tools
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]