Read Memories
Searches past Claude Code session logs to recover decisions, patterns, and context
Category: development Source: duckdb/duckdb-skillsWhat Is This?
Overview
Read Memories is a development skill that enables Claude Code to search through past session logs and recover contextual information from previous work sessions. When invoked, it queries stored session data to surface past decisions, established patterns, and unresolved tasks that may be relevant to the current work. This allows the assistant to operate with continuity across sessions rather than starting from scratch each time.
The skill works by running a silent background query against accumulated session logs, absorbing the results, and applying that recovered context to the current task. The process is transparent to the user in the sense that it does not interrupt the workflow with narration or status updates. The assistant simply becomes more informed before proceeding.
This approach addresses one of the most common friction points in AI-assisted development: the loss of context between sessions. By maintaining a searchable record of past interactions and decisions, Read Memories allows development workflows to build on prior work rather than repeatedly re-establishing the same ground.
Who Should Use This
- Developers working on long-running projects who need continuity across multiple Claude Code sessions
- Engineers who make architectural decisions during sessions and want those decisions respected in future work
- Teams where multiple contributors use Claude Code and need shared context about project conventions
- Developers who frequently context-switch between projects and need to quickly re-orient when returning to a codebase
- Anyone who has experienced the frustration of re-explaining the same project constraints or preferences in every new session
Why Use It?
Problems It Solves
- Context loss between sessions forces developers to repeatedly re-explain project structure, naming conventions, and prior decisions, wasting time and introducing inconsistency.
- Without memory of past work, the assistant may suggest approaches that were already considered and rejected, leading to circular discussions.
- Unresolved tasks and known issues can be forgotten between sessions, causing work to stall or duplicate effort.
- Developers working across multiple projects struggle to maintain separate context for each codebase without a reliable retrieval mechanism.
- Onboarding to a project mid-stream is difficult when there is no record of the decisions that shaped the current state of the code.
Core Highlights
- Searches past Claude Code session logs using keyword-based queries
- Supports scoping queries to the current project using the
--hereflag - Operates silently without interrupting the development workflow
- Recovers decisions, patterns, and unresolved work items from prior sessions
- Works across all projects or can be narrowed to a specific context
- Integrates with the Bash tool for direct log querying
- Proactive invocation allows the assistant to self-enrich before responding
How to Use It?
Basic Usage
Invoke the skill with a keyword to search across all past session logs:
read-memories <keyword>
To scope the search to the current project directory:
read-memories <keyword> --here
For example, to recover context about database schema decisions:
read-memories schema --here
Specific Scenarios
Recovering architectural decisions: When starting work on a module that was discussed in a previous session, invoke the skill with a relevant term before making changes. The assistant will silently retrieve prior context about design choices and apply them without prompting.
Resuming unfinished work: If a session ended before a task was completed, use a keyword related to that task to surface the last known state and any notes about what remained to be done.
Real-World Examples
A developer returns to a DuckDB extension project after two weeks. Before writing new query logic, the assistant runs read-memories query-optimization --here and recovers notes about a specific indexing approach that was agreed upon in a prior session.
A team member picks up work on an API layer. The assistant queries read-memories authentication and finds a prior decision to use token-based auth with a specific expiry policy, avoiding a redundant discussion.
When to Use It?
Use Cases
- Resuming development on a project after a break
- Enforcing consistency with previously established coding conventions
- Avoiding re-discussion of already-resolved architectural questions
- Tracking down the rationale behind an existing implementation choice
- Coordinating work across multiple contributors using the same assistant
- Recovering the state of a debugging session that was interrupted
- Applying project-specific patterns when generating new code
Important Notes
Requirements
- Claude Code must have been used in prior sessions for logs to exist
- The Bash tool must be available and permitted in the current environment
- Session logs must be stored in a location accessible to the skill query