Claude Code Plugin

The official Snipara plugin for Claude Code adds slash commands, auto-invoked skills, and automation hooks that make context optimization and safe code execution seamless.

MCP Server Required

The plugin adds commands and skills on top of the Snipara MCP server. You need the MCP server configured for the plugin to work.

What the Plugin Adds

14 Slash Commands

Invoke workflows, search, memory, and execution directly from Claude Code.

/snipara:lite-mode/snipara:full-mode/snipara:remember/snipara:recall+10 more

5 Auto-Invoked Skills

Claude automatically queries docs, recalls context, and plans tasks.

✓ query-docs
✓ recall-context
✓ plan-task
✓ execute-code
✓ chunk-implement

Automation Hooks

Automatic context recall at session start and memory save on edits.

✓ SessionStart auto-recall
✓ PostToolUse auto-remember
✓ Commit memory on push

Installation

Step 1: Configure MCP Server

Set up the Snipara MCP server first. See the Claude Code Integration guide.

Step 2: Install the Plugin

Option A: From Marketplace (recommended)

/plugin marketplace add Snipara/snipara-claude

/plugin install snipara@snipara-plugins

Option B: Clone and load locally

git clone https://github.com/Snipara/snipara-claude.git

claude --plugin-dir ./snipara-claude

Step 3: Verify

/help

snipara:lite-mode, snipara:full-mode, snipara:remember, snipara:recall ...

Slash Commands Reference

Command	Description	Example
Workflows
`/snipara:lite-mode`	Quick bug fixes, small features (<5 files)	`/snipara:lite-mode auth timeout`
`/snipara:full-mode`	Complex features with RLM Runtime (5+ files)	`/snipara:full-mode OAuth`
Documentation
`/snipara:search`	Search docs with regex pattern	`/snipara:search auth`
`/snipara:team-search`	Search across all team projects	`/snipara:team-search rate limit`
Memory
`/snipara:remember`	Save context, decisions, or learnings	`/snipara:remember Chose JWT`
`/snipara:recall`	Search previous memories	`/snipara:recall OAuth`
Planning
`/snipara:plan`	Generate execution plan for complex tasks	`/snipara:plan user auth`
`/snipara:decompose`	Break task into sub-queries	`/snipara:decompose OAuth`
Team
`/snipara:shared`	Get team coding standards	`/snipara:shared`
`/snipara:inject`	Set session context for all queries	`/snipara:inject auth refactor`
RLM Runtime
`/snipara:run`	Execute code locally with RLM	`/snipara:run parse logs`
`/snipara:docker`	Execute code in Docker isolation	`/snipara:docker run pytest`
`/snipara:visualize`	Launch trajectory dashboard	`/snipara:visualize`
`/snipara:logs`	View RLM execution logs	`/snipara:logs`

Workflows

/snipara:lite-mode

Quick bug fixes, small features (<5 files)

/snipara:full-mode

Complex features with RLM Runtime (5+ files)

Documentation

/snipara:search

Search docs with regex pattern

/snipara:team-search

Search across all team projects

Memory

/snipara:remember

Save context, decisions, or learnings

/snipara:recall

Search previous memories

Planning

/snipara:plan

Generate execution plan

/snipara:decompose

Break task into sub-queries

Team

/snipara:shared

Get team coding standards

/snipara:inject

Set session context

RLM Runtime

/snipara:run

Execute code locally

/snipara:docker

Execute in Docker isolation

/snipara:visualize

Launch trajectory dashboard

/snipara:logs

View execution logs

Auto-Invoked Skills

Skills are automatically triggered by Claude when they match the user's intent. No manual invocation needed.

query-docs

Triggers when: You ask about documentation, APIs, or architecture

Calls rlm_context_query with semantic search

recall-context

Triggers when: You reference previous decisions or resume work

Calls rlm_recall to find relevant memories

plan-task

Triggers when: You request a complex multi-file implementation

Calls rlm_plan and rlm_decompose

execute-code

Triggers when: You ask to run tests or validate code safely

Runs code via RLM Runtime with Docker

chunk-implement

Triggers when: Complex feature spanning 5+ files

Queries context per chunk, executes, remembers

RLM Runtime Integration

The plugin optionally integrates with RLM Runtime for safe code execution in Docker containers.

✓

Isolated Execution

Run code in Docker containers without host access

✓

Trajectory Logging

Full execution trace for debugging and review

✓

Multi-Provider

Works with OpenAI, Anthropic, and LiteLLM

✓

Visualization

Interactive Streamlit dashboard for analysis

Setup RLM Runtime

pip install rlm-runtime[all]

rlm init

rlm run --env docker "print('Hello from RLM')"

Optional Component

RLM Runtime requires Docker and a separate LLM API key (Anthropic or OpenAI). The plugin's Snipara features (commands, memory, search) work without RLM Runtime.

Workflow Examples

LITE Mode: Quick Bug Fix

/snipara:lite-mode authentication timeout bug

[rlm_context_query: "authentication timeout"]

Found: src/auth/middleware.ts - timeout at 5000ms

Fix applied. Tests passing.

FULL Mode: Complex Feature

/snipara:full-mode implement OAuth integration

Phase 1: Loading team standards...

Phase 2: Generating plan (6 chunks)...

Phase 3: Implementing chunk 1/6

Phase 4: Running tests...

All 6 chunks implemented. 47 tests passing.

Memory-Driven Development

Session 1: Working on registration

/snipara:remember "Completed registration API"

Memory saved.

Session 2 (next day):

[Auto-recall] Last: registration API, next: email

Source Code & Resources

GitHub Repository

Source code, issues, and releases.

MCP Server Setup

Configure the MCP server that powers the plugin.

Next Steps

MCP Tools Reference

Full documentation for all 43 Snipara MCP tools.

RLM Runtime

Safe code execution with Docker isolation and trajectory logging.

Automation Hooks

Context preservation and automatic memory across sessions.

Agent Memory

Remember decisions and context across sessions.