External Memory Plugins
Builds, tests and troubleshoots memory provider plugins for the Nexent agent platform.
Install
npx skills add ModelEngine-Group/nexent --skill external-memory-plugin -g -a claude-code -y- Repository
- ModelEngine-Group/nexent6.1k
- Category
- AI and agents
- Updated
- 4d ago
- License
- MIT
- Context cost
- Lightabout 1,050 tokens
- Runs scripts
- No, instructions only
What it does
Develop, adapt, review, test, and troubleshoot Nexent external memory provider plugins, including plugin.yaml manifests, searchable and ingestible provider protocols, error mapping, network-isolated unit tests, deployment configuration, and Mem0-based examples. Use when adding a new external memory vendor, changing an existing memory plugin, validating partner adapters, or diagnosing plugin discovery, authentication, retrieval, ingestion, or CI failures.
External Memory Plugin
Develop external memory adapters against the current Nexent contract and prove them without live-service CI dependencies.
Establish scope
- Read the repository
AGENTS.mdand invoke its SPEC workflow for production behavior changes. - Inspect, rather than assume, these current contracts:
sdk/nexent/memory/providers/base.pysdk/nexent/memory/models.pybackend/services/memory_provider_plugin_loader.pybackend/services/memory_external_provider_service.py
- Read references/plugin-contract.md before creating or reviewing a manifest or provider.
- Read references/mem0-example.md when implementing HTTP transport, error mapping, or tests.
- Preserve tenant, user, Agent, and conversation scope. Never print or persist credentials.
Implement the adapter
- Create
<plugin-dir>/<name>/plugin.yamlandprovider.py. - Declare only implemented capabilities in
implements. - Make protocol methods asynchronous and return the exact Nexent models.
- Honor provider-level
timeout_secondsand result limits. - Map remote failures to
ProviderErrorand the retry exception matching recovery semantics. - Keep OTel attributes low-cardinality and free of queries, memory content, user identifiers, tenant identifiers, and secrets. Rely on the orchestration service's standard instrumentation unless provider-specific spans add actionable detail.
- Treat the deployment search/ingest switches as kill switches and the provider
enabledfield as the instance switch; do not bypass them in normal runtime paths.
