Give any AI coding agent governed project memory across sessions, tools, and machines.
Agent-Memory-Skills is the official companion skill repository for Agent Memory Bridge (AMB).
While AMB provides the durable authority (SQLite/WAL, FTS5 lexical recall, Streamable HTTP / stdio MCP server), these skills give your LLM the exact procedural knowledge to connect, configure, and use memory effectively.
Connecting a host and using memory day-to-day have fundamentally different lifecycles and context footprints:
┌────────────────────────────────────────┐
│ User / Agent Prompt │
└──────────────────┬─────────────────────┘
│
┌─────────────────────────┴────────────────────────┐
▼ ▼
"Connect AMB to Codex / OpenCode" "Remember this architectural decision"
"Deploy remote Docker authority" "Recall prior gotchas for this bug"
"Configure tool allowlists / hooks" "Revise outdated database constraint"
│ │
▼ ▼
┌─────────────────────┐ ┌─────────────────────┐
│ amb-connect │ │ amb │
│ (Host Integration) │ │ (Everyday Memory) │
└──────────┬──────────┘ └──────────┬──────────┘
│ │
▼ ▼
• Discovers host capabilities • Resolves bound project scope
• Connects stdio or HTTP MCP • Contextual recall (<= 5 results)
• Sets 6-tool Everyday profile • Checks live code vs memory WHY
• Wires lifecycle hooks (optional) • Stores confirmed decisions
• Proves 2-session First Win • Revises/annotates existing rows
| Skill | Primary Job | Typical Triggers |
|---|---|---|
skills/amb |
Everyday Memory | "Remember that we use WAL mode", "What did we decide about auth?", "That database rule changed; update it." |
skills/amb-connect |
Connection & Integration | "Connect AMB to OpenCode", "Set up remote HTTP memory on our server", "Why is AMB reporting missing tools?" |
AMB exposes exactly 17 public MCP tools backed by schema v12. Rather than overwhelming models with 17 simultaneous tool schemas, Agent-Memory-Skills establishes a recommended 6-tool Everyday Profile for standard coding sessions:
| Profile | Tools Included | When to Use |
|---|---|---|
| Everyday Memory (Recommended) | recall, browse, stats, store, revise, annotate |
Standard coding, bugfixes, architectural decisions, and memory maintenance. |
| Retrieval Feedback (Optional) | feedback |
Explicit retrieval evaluation using 15-minute signed recall receipts. |
| Maintenance (Operator) | forget, promote, export |
Explicit record deletion, record promotion, or portable namespace snapshot. |
| Tracked Work (Advanced) | begin_run, record_run_event, get_run, complete_run |
Governed execution runs with server-minted verification receipts. |
| Coordination Signals (Advanced) | claim_signal, extend_signal_lease, ack_signal |
Multi-agent handoffs, claim leases, and task polling. |
Load amb-connect when setting up a new agent host or server:
user> Connect AMB to my coding environment.
agent> [Loads amb-connect]
1. Discovers host capabilities (stdio subprocess vs HTTP MCP).
2. Connects to existing AMB or provisions new instance.
3. Recommends the 6-tool Everyday Profile.
4. Verifies via server doctor and the Two-Session First Win round trip.
Once connected, load amb during coding sessions:
user> Remember that we only merge pull requests after CI passes, because broken main blocked releases twice this week.
agent> [Loads amb]
Stores explicit decision under bound project namespace with claim, reason, and scope.
--- [Fresh session later] ---
user> What is required before we merge a PR?
agent> [Loads amb]
Recalls stored decision: CI must pass before merging.
AMB runs over standard MCP. It connects to compatible agent runtimes capable of running MCP stdio or Streamable HTTP meeting its authentication requirements:
- Local IDEs & CLIs: OpenCode, Codex CLI, Claude Desktop, Cursor, Cline, Windsurf.
- Server Bots & Custom Agents: Grok bots, Discord/Slack agents, Dots, custom Python/Node scripts.
- Remote / NAS Authorities: Dockerized AMB serving Streamable HTTP at
/mcpbehind bearer token authentication.
Agent-Memory-Skills/
├── skills/
│ ├── amb/ # Everyday memory skill
│ │ ├── SKILL.md # Core skill entrypoint & 5-step loop
│ │ └── references/
│ │ ├── everyday_workflows.md # Step-by-step memory patterns
│ │ ├── record_templates.md # Canonical decision, gotcha, constraint templates
│ │ ├── revision_and_correction.md # revise vs annotate vs forget
│ │ ├── scope_and_namespaces.md # project:<slug> vs domain:<name>
│ │ ├── tool_matrix.md # Complete 17-tool contract breakdown
│ │ └── advanced_modes.md # Runs ledger and coordination signals
│ │
│ └── amb-connect/ # Host connection & transport skill
│ ├── SKILL.md # 5-stage connection protocol
│ └── references/
│ ├── capability_discovery.md # Detecting host capabilities
│ ├── transports_and_auth.md # stdio vs HTTP bearer auth
│ ├── local_stdio_recipe.md # Setting up local stdio MCP
│ ├── remote_http_recipe.md # Connecting to remote HTTP MCP
│ ├── docker_deployment.md # Official Docker Compose deployment
│ ├── tool_profiles.md # Everyday (6) vs Full (17) filtering
│ ├── lifecycle_and_hooks.md # Generic lifecycle-hook automation
│ └── verification_and_repair.md # Doctor, verify, and first-win test
│
├── scripts/
│ └── validate_skills.py # Integrity, link, and contract validator
├── tests/
│ ├── test_skills_structure.py # Structural test suite
│ └── test_contracts.py # Technical contract test suite
└── .github/workflows/
└── ci.yml # Automated GitHub Actions test runner
Run the included test and validation suites:
# Validate skill markdown, frontmatter, references, and tool matrix
python3 scripts/validate_skills.py
# Run unit tests
python3 -m unittest discover -s testsMIT License. See LICENSE.