session-memory
Enables cross-session context persistence for Claude Code, managing history, project decisions, and workflow patterns to ensure seamless task continuation.
Introduction
The Session Memory skill provides a robust architectural layer for maintaining project continuity across independent Claude Code sessions. Designed for developers and engineers, it acts as a persistent knowledge bridge that tracks project evolution, architectural decisions, and recurrent technical patterns. By utilizing a structured directory at .claude/memory/ and integrating with the internal Agent Trace system, it ensures that your workspace history is never lost between CLI invocations, allowing for more efficient, long-term development cycles.
-
Automatically captures session-level logs using unique session IDs for high-fidelity traceability of past tasks and file modifications.
-
Maintains an SSOT (Single Source of Truth) for project-wide decisions and technical patterns, helping to avoid repeated implementation errors and enforce architectural standards.
-
Integrates deeply with context.json to store high-level project metadata, technical stacks, and current development phases, enabling rapid environment recovery.
-
Supports advanced commands like /recap for immediate synthesis of current session progress and --resume for restoring complex, long-running workflows.
-
Implements a sophisticated 3-layer memory architecture that distinguishes between implicit automatic learning, project-specific manual knowledge, and specialized agent-task tracking.
-
Use this skill when prompted by queries like 'what did we do last time?' or 'continue from before' to restore work context.
-
Avoid loading this skill for transient, ad-hoc information requests or routine code reviews where session persistence is not required.
-
Adhere to the defined MEMORY_POLICY.md to decide which files should be tracked via Git (decisions.md/patterns.md) and which should remain local to the development machine (session-log.md/context.json).
-
Leverage the provided bash snippets and agent-trace analysis tools to manually extract complex history if the automated summary requires fine-tuning.
-
Always be mindful of privacy constraints by ensuring that sensitive tokens or credentials are not recorded in the plaintext markdown logs used for context persistence.
Repository Stats
- Stars
- 633
- Forks
- 67
- Open Issues
- 7
- Language
- Shell
- Default Branch
- main
- Sync Status
- Idle
- Last Synced
- Apr 29, 2026, 06:25 AM