triple-memory

What This Skill Does

The Triple Memory system is a sophisticated, multi-layered cognitive architecture designed for OpenClaw agents that require long-term context retention. It functions by synchronizing three distinct memory backends: LanceDB for semantic auto-recall of past conversations, Git-Notes for structured, branch-aware decision tracking, and a file-based search engine for deep workspace knowledge. By operating these three systems in parallel, the skill ensures that information is not only captured automatically as preferences or facts during a chat but is also persisted into human-readable files and git-tracked notes. This prevents the 'memory drift' often associated with AI models, allowing the agent to maintain high-fidelity context across long-running development sessions or complex, multi-day projects.

Installation

Installation follows a multi-step configuration process. First, ensure you have the prerequisite tools by running clawdhub install git-notes-memory and placing the scripts/file-search.sh utility in your root workspace. Next, integrate the LanceDB plugin by adding the provider configuration into your agent settings, ensuring an OpenAI API key is provided for the embedding model. Finally, append the memory-flush configuration to your agent's compaction settings. This critical final step enables the 'safeguard' mode, which triggers an automated context dump to your memory/ directory once your token usage hits the 80% threshold, ensuring zero information loss during model pruning.

Use Cases

This skill is designed for complex engineering workflows where persistent state is mandatory. Primary use cases include tracking architectural decisions across long-term codebases, managing shifting user requirements during product development, and maintaining a 'knowledge graph' of team-specific preferences. It is particularly effective for developers needing to jump between branches, as the Git-Notes component automatically pivots memory context based on the current git state.

Example Prompts

1. "Remember that my preferred database for this project is PostgreSQL because the team has extensive experience with it."
2. "Look through the project memory and the memory/2023-10-12.md file to summarize the key decisions we made regarding our authentication flow."
3. "I need to find the specific constraint we decided on for the API rate limiting; search the git-notes and workspace files for this."

Tips & Limitations

To maximize effectiveness, always run the sync command at the start of every session to align the agent with the current repository state. Be mindful of token usage; while the auto-flush is powerful, ensure your memory directory is included in your .gitignore if you do not want to commit persistent agent logs to your main code repository. Note that LanceDB relies on remote embedding APIs, so ensure you have a stable network connection when initiating or closing high-context tasks.