tracemem-plugins

TraceMem Plugin for Claude Code

Decision memory for Claude Code. Automatically tracks architectural decisions, code changes, and engineering choices into TraceMem's governed audit trail.

Quickstart

1. Install the plugin

From GitHub (recommended):

/plugin marketplace add tracemem/tracemem-claude-code-plugin
/plugin install tracemem@tracemem-plugins

Or install directly from GitHub without adding the marketplace:

claude plugin install tracemem@tracemem-plugins --marketplace-source github:tracemem/tracemem-claude-code-plugin

From a local clone:

git clone https://github.com/tracemem/tracemem-claude-code-plugin.git

Then in Claude Code:

/plugin install /path/to/tracemem-claude-code-plugin

2. Set your API key

# Add to ~/.bashrc, ~/.zshrc, or your shell profile
export TRACEMEM_API_KEY=your_api_key_here

Or in your project's .env file:

TRACEMEM_API_KEY=your_api_key_here

Get your API key from app.tracemem.com under your project settings.

3. Verify setup

/tracemem-doctor

This checks your API key, MCP server connectivity, and available tools.

Available Commands

Command	Description
/tracemem-open [action]	Open a decision envelope for the current task
/tracemem-close [commit|abort]	Close the active decision envelope
/tracemem-note <message>	Add a note to the active decision
/tracemem-status	Show the active decision state
/tracemem-trace [id]	View the decision trace
/tracemem-receipt [id]	Get the immutable receipt for a closed decision
/tracemem-record <description>	Record a one-shot architecture/engineering decision
/tracemem-search [query]	Search past decisions for precedent
/tracemem-doctor	Verify plugin setup and connectivity

Actions for /tracemem-open

Action	Intent	Use when
edit_files	code.change.apply	General code editing (default)
refactor	code.refactor.execute	Restructuring code
run_command	ops.command.execute	Operational commands
deploy	deploy.release.execute	Deployments and releases
secrets	secrets.change.propose	Secret modifications
db_change	data.change.apply	Database changes
review	code.review.assist	Code review

How Automatic Tracking Works

The plugin uses three layers to ensure decisions are captured:

Skills

The decision-tracking skill teaches Claude when and how to track decisions autonomously. When Claude detects significant technical choices (architecture, dependencies, schema changes, API modifications), it records them automatically.

Hooks

Lifecycle hooks provide guaranteed automation:

• Session Start: Injects TraceMem context so Claude knows tracking is available
• Post-Tool Use: After file writes or bash commands, evaluates whether the change represents a significant decision
• Stop: Before finishing, reviews the session for missed decisions and auto-closes open envelopes
• Session End: Lightweight cleanup logging

Agents

Two specialized agents handle decision workflows:

• Decision Tracker: Identifies and records key decisions during coding
• Decision Reviewer: Reviews the decision trace for completeness before closing

Two Recording Paths

One-shot recording (/tracemem-record)

For architecture and engineering decisions that don't require data product operations. This is the most common path during coding:

/tracemem-record Use Redis for session caching

Claude will search for existing decisions, gather details from context, and record the decision in a single operation.

Full envelope (/tracemem-open ... /tracemem-close)

For decisions involving governed data reads/writes, policy evaluations, or approval workflows:

/tracemem-open edit_files
... work on the task ...
/tracemem-close

Configuration

Environment Variables

Variable	Required	Description
TRACEMEM_API_KEY	Yes	API key for TraceMem authentication
TRACEMEM_MCP_URL	No	Override MCP server URL (default: https://mcp.tracemem.com)

MCP Server

The plugin connects to TraceMem's cloud-hosted MCP server via HTTP. No local process to manage. The connection is configured in .mcp.json and started automatically when the plugin is enabled.

What Gets Tracked

Always tracked:

• Database schema changes and migrations
• New dependency additions
• API contract changes
• Infrastructure and deployment decisions
• Security-relevant choices
• Explicit trade-offs discussed with the user

Never tracked:

• Formatting, typo fixes, comment additions
• Routine test runs
• File navigation or code reading
• Clarifying questions

Security