Terminal.skills
Skills/claude-subconscious
>

claude-subconscious

Add persistent subconscious memory to Claude Code that survives across sessions — automatically stores decisions, patterns, and context. Use when: maintaining continuity across Claude Code sessions, building agents with long-term project memory, avoiding repeated context setup.

#claude-code#memory#persistence#context#letta
terminal-skillsv1.0.0
Works with:claude-codeopenai-codexgemini-clicursor
Source

Usage

$
✓ Installed claude-subconscious v1.0.0

Getting Started

  1. Install the skill using the command above
  2. Open your AI coding agent (Claude Code, Codex, Gemini CLI, or Cursor)
  3. Reference the skill in your prompt
  4. The AI will use the skill's capabilities automatically

Example Prompts

  • "Review the open pull requests and summarize what needs attention"
  • "Generate a changelog from the last 20 commits on the main branch"

Information

Version
1.0.0
Author
terminal-skills
Category
Development
License
MIT

Use Cases

→Optimize AI Coding Agent for Production Team Use

Documentation

Overview

Add a persistent memory layer to Claude Code using Letta. A background agent watches your sessions, reads your files, builds up memory over time, and whispers guidance back — so Claude never forgets what you've worked on.

Source: letta-ai/claude-subconscious (2.5k+ stars)

Claude Code forgets everything between sessions. Claude Subconscious is a second agent running underneath — watching, learning, and whispering back. It uses Letta's Conversations feature so a single agent serves multiple sessions with shared memory.

Instructions

1. Install the plugin

From Claude Code Plugin Marketplace:

bash
/plugin marketplace add letta-ai/claude-subconscious
/plugin install claude-subconscious@claude-subconscious

From source:

bash
git clone https://github.com/letta-ai/claude-subconscious.git
cd claude-subconscious
npm install
/plugin enable .

2. Configure Letta API key

bash
export LETTA_API_KEY="your-api-key"

Get your API key from app.letta.com.

3. Optional configuration

bash
# Memory mode: "whisper" (default), "full" (blocks + messages), "off" (disable)
export LETTA_MODE="whisper"

# Pin to a specific agent (auto-created if not set)
export LETTA_AGENT_ID="agent-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"

# Custom Letta server (for self-hosted)
export LETTA_BASE_URL="https://your-letta-server.com"
ModeBehaviorToken Cost
whisperShort guidance before each promptLow
fullFull memory blocks + message history injectedHigher
offPlugin disabledNone

4. Self-hosted option

bash
pip install letta
letta server --port 8283
export LETTA_BASE_URL="http://localhost:8283"

Examples

Example 1: Architecture decision memory

After discussing a REST-to-GraphQL migration in one session, you start a new session the next day. The subconscious whispers:

[subconscious] Last session you decided to switch from REST to GraphQL for the
user service. The migration is 60% done — resolvers for User and Project are
complete, Order and Payment still need conversion. You preferred code-first
schema generation with TypeGraphQL.

You can immediately continue the migration without re-explaining context.

Example 2: Pattern and convention recall

After establishing naming conventions and test patterns across several sessions, you start working on a new module. The subconscious whispers:

[subconscious] This project uses barrel exports (index.ts) for all modules.
Tests follow the pattern: describe('[ModuleName]') with 'should' prefixed
test names. You prefer integration tests over unit tests for API routes.
Database fixtures go in tests/fixtures/.

The new module code follows established patterns without you having to look up past decisions.

Guidelines

  • Start fresh per project — Each project benefits from its own agent (auto-created by default)
  • Whisper mode is enough — full mode uses more tokens; whisper gives 90% of the value
  • Pair with CLAUDE.md — Use CLAUDE.md for static project context, subconscious for dynamic memory
  • Check memory — Visit app.letta.com to see what your agent has learned
  • Linux tmpfs workaround — If installation fails with EXDEV: cross-device link not permitted, set export TMPDIR="$HOME/.claude/tmp"
  • Requires a Letta API key (free tier available) or self-hosted Letta server
  • Background processing adds ~1-2s latency per whisper
  • Memory quality depends on session length — short sessions produce less useful memories

References