The Problem
LLM coding assistants like Claude Code are stateless by default. Each session starts with no memory of prior work, forcing users to re-explain project context, architectural decisions, and their own preferences. For long-running projects or teams, this repetition wastes tokens and produces inconsistent output.
What This Does
claude-supermemory is a Claude Code plugin that persists conversation context to the Supermemory service. On session start, src/context-hook.js fetches relevant memories and injects them into Claude's context. During the session, src/observation-hook.js captures conversation turns and src/add-memory.js stores them for future retrieval.
The plugin also includes a codebase indexing command (src/add-memory.js via the /claude-supermemory:index command) and a super-search skill (plugin/skills/super-search/SKILL.md) that lets the agent query past work on demand. All hooks are wired through plugin/hooks/hooks.json, and the plugin communicates with Supermemory via src/lib/supermemory-client.js.
How To Use It
Setup: Install the plugin from the marketplace or a local directory, then set the required API key. The README documents these commands verbatim:
/plugin marketplace add supermemoryai/claude-supermemory /plugin install claude-supermemory export SUPERMEMORYCCAPIKEY="sm..."
Configuration: Optional environment variables (SUPERMEMORYSKIPTOOLS, SUPERMEMORY_DEBUG) and a settings file at ~/.supermemory-claude/settings.json control which tools are captured and debug logging. The src/lib/settings.js module handles this.
Running it: Once installed, the plugin activates automatically when Claude Code starts. Use /claude-supermemory:index to index the current codebase, and /claude-supermemory:logout to clear credentials.
Real-World Use
A team maintains a monorepo with multiple services. A developer starts a session to fix a bug in the auth service. The plugin injects memories from previous sessions: the team prefers TypeScript, uses Bun, and recently refactored the auth flow. The developer asks "how did we solve the token refresh issue last time?" and the super-search skill retrieves the relevant context. The fix is written in one pass instead of three sessions of rediscovery.
Code Health & Issues
Med - No test files detected - The repository has zero test files. The hook scripts (src/context-hook.js, src/observation-hook.js) handle external API calls and file I/O, which are exactly the paths that need tests. This is a significant gap for a plugin that runs inside a developer's editor. Med - No LICENSE file - The README states MIT, but no license file exists in the repo. This creates ambiguity for downstream users. Low - External API dependency - The plugin requires a paid Supermemory Pro subscription (src/lib/supermemory-client.js). If the service has downtime or rate limits, the plugin's failure handling is untested. Low - Credential handling - The API key is stored in an environment variable, which is standard, but the auth flow in src/lib/auth.js and the HTML templates (src/templates/auth-error.html, src/templates/auth-success.html) suggest a browser-based OAuth flow that adds complexity.
The Bottom Line
This is a well-structured plugin with clear separation of concerns across its hook scripts and library modules. The core value—persistent context across sessions—is real and useful for anyone using Claude Code daily. The lack of tests and missing license file are the main concerns; the paid service dependency means it's not a drop-in solution for everyone. Best suited for teams already using Supermemory or willing to pay for it.