Agent Shared Mem — Development skill for Claude Code
Shared global and project memory runtime for Codex, Claude Code, OpenCode, and Hermes Agent.
How to install Agent Shared Mem
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open d4em0n/agent-shared-mem and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Agent Shared Mem does
Shared global and project memory runtime for Codex, Claude Code, OpenCode, and Hermes Agent.
Alternatives in Development
- Petdex — A public gallery of animated pets for Codex, Claude Code, DeepSeek Harness, Hermes, OpenCode, Gemini CLI, and 4k ★
- Agentic Stack — One brain, many harnesses 2.2k ★
- Project Memory: AgentSys — Cross-tool compatible: Works across all 5 platforms - Claude Code, OpenCode, Codex CLI, Cursor, and Kiro 622 ★
README
shared-memory-runtime v2
A tool-neutral persistent-memory runtime for Codex, Claude Code, OpenCode, and Hermes Agent.
Storage
Version 2 has exactly two scopes:
~/.local/share/agent-memory/global/ global preferences and machine facts
/.agent-memory/ repository-specific facts
Each store contains `MEMORY.md`, `PROTOCOL.md`, `SCOPE.md`, and one Markdown file per fact. Project facts override same-title global facts. A store may use `INDEX-.md` files to keep the root index small; prompt recall follows those indexes recursively while SessionStart injects only `MEMORY.md`.
Index entries always point to fact files inside the memory store. A fact holds the concise durable result and may link to authoritative project artifacts. Updating another ledger or documentation tree does not replace shared-memory maintenance, and agents do not need an explicit user request to update this store when their work changes durable knowledge.
Project stores are added to `.git/info/exclude`, keeping them local without editing the repository's committed `.gitignore`.
Initialize
shared-memory init global
shared-memory init project --project /path/to/repo
shared-memory status --project /path/to/repo
`status` reports root-index bytes, nested indexes, indexed facts, invalid or duplicate pointers, and the estimated combined SessionStart context health. The runtime prepends an index-maintenance notice before an oversized index, so the notice remains visible even if a client truncates the index itself.
Default warning thresholds are 14 KiB per root index and 16 KiB for the combined injected context. Index splitting is semantic and remains agent-led: move coherent pointer lines into an `INDEX-.md`, leave one descriptive root pointer, and never move or duplicate the underlying fact files. Thresholds can be adjusted with:
AGENT_MEMORY_INDEX_WARN_BYTES
AGENT_MEMORY_INDEX_MAX_BYTES
AGENT_MEMORY_CONTEXT_WARN_BYTES
AGE
Related Skills
Unified Agent Memory
Unified fleet-wide agent memory system for DeepSeek Harness — shared Obsidian Vault for dsh, Codex, Claude Cod
Armor Memory
ARMOR Minimal Stable is a deterministic shared memory architecture for Codex, Claude Code, and Hermes Agent.
All Agent Shared Memory
Zero-dependency cross-agent memory + session bridge for Claude Code, Codex, opencode and ZCode: one auditable
Agent Hop
Runtime for coding-agent harnesses. Run Claude Code, Codex, OpenCode, Pi, and Grok in one terminal; hop live b
Windows Agent Orchestrator
Windows-native, headless orchestrator for dispatching certified worker agents through local agent runtimes (cl
Agnt
Control coding-agent CLIs from your iPhone — Codex, Claude Code, opencode, and any other agent you plug in. ag
Related Agents
Ensemble Curator
High-volume triage agent for external-AI-agent PRs (Codex, Jules, Hermes, Droid, Aider, etc.). Analyzes, conso
Mer Recall
Worker that loads a project's Memory v2 context (mem_state, mem_context, mem_search, mem_get) and reconciles i
Maker
Reviews OpenCode sessions (shared links or local history), finds repeatable workflows, and proposes automation