v0.6.3

MCP domain failures now remain readable in clients that validate structured outputs against success schemas. Errors preserve their code, message, and details as JSON text with isError: true; successful results still include typed structuredContent.

Claim tool schemas explain kind:name resources, and retrieval descriptions call out full-record expansion before consequential use. Pair with Agent Memory plugin 0.1.1.

Validation: 132 application tests, strict typing and formatting, baseline guidance verification, native standalone binary verification, and fresh Codex, Claude, Grok, and Cursor coding trials. Each host retrieved the full relevant memory, claimed before editing, released its claim, and passed the unchanged scheduler tests.

Agent Memory plugin 0.1.1

Requires memd 0.6.3 or later.

• Correct Cursor headless permissions to use the plugin server name plugin-agent-memory-agent-memory.
• Make claim acquisition, resource naming, and recovery from failed acquisition explicit in the portable skill.
• Require full-record reads before consequential use of truncated evidence.
• Document Grok installation-source overlap and verify selected sources during upgrades.

Fresh Codex, Claude, Grok, and Cursor trials passed an additional scheduler case with full-record reads, claims before edits, and no claims left behind. These trials used the documented project instruction; plugin installation alone still does not guarantee automatic retrieval. Codex's candidate test used project skill discovery; the other hosts used the plugin bundle. Guidance remains repository agnostic.

Agent Memory plugin 0.1.0

Agent Memory now ships one repository-independent skill with plugin adapters for Codex, Claude Code, Grok Build and Cursor. Other Agent Skills hosts can use the standalone skill archive with their own MCP configuration.

• Native manifests, portable Agent Plugins files, and Git marketplace catalogs share one maintained skill.
• Plugin and standalone skill archives include SHA-256 checksums.
• Requires memd 0.6.2 or later running on the agent's machine. Installing the plugin does not install or restart the daemon.
• Plugin versions are independent of daemon versions; this release does not change the latest memd release.

Installation and update instructions

Validation: 130 tests / 921 assertions, strict typing, formatting, guidance checks, native manifest validation and portable schema validation passed. Live trials loaded the packaged skill and retrieved memory in Codex CLI 0.157.1, Claude Code 2.1.283, Grok Build 1.0.40 and Cursor CLI 2026.09.08-6caf4ff. Codex and Claude used the bundled MCP connection; Grok and Cursor used their existing agent-memory connections. These were scoped retrieval trials, not coding-accuracy benchmarks.

v0.6.2

Codex CLI can now connect to memd using MCP 2025-06-18. MCP 2026-07-28 remains supported; both HTTP and stdio use the same tools and services through the official SDK.

The retrieval evaluator now compares full, compact and briefing results, checks whether qualifications survive truncation, and detects changed record versions. Shared guidance describes a repository-independent workflow for maintaining and evaluating memory.

Validation covers both protocol versions, text and structured results, conflicts, project isolation, and standalone offline binaries.

v0.6.1

What's Changed

• Fix router path shadowing and over-long query embeddings; release 0.6.1 by @rhymiz in #1

New Contributors

• @rhymiz made their first contribution in #1

Full Changelog: v0.6.0...v0.6.1

v0.6.0

Full Changelog: https://github.com/rhymiz/agent-memory/commits/v0.6.0