agent-memory

The agent-memory skill provides a robust, persistent storage mechanism designed to help AI agents maintain context over extended development lifecycles. By establishing a dedicated directory at .claude/skills/agent-memory/memories/, this skill allows agents to save valuable research, architectural rationales, complex problem-solving steps, and in-progress development tasks that would otherwise be lost when a chat session terminates. It functions as a structured knowledge base that agents can proactively query before starting new tasks, ensuring technical consistency and reducing redundant investigation.

• Persistent storage of project-specific insights, including 'gotchas' in the codebase and complex dependency resolution patterns.

• Support for categorized organization using folders and YAML frontmatter to manage metadata like status, tags, and update timestamps.

• Command-line integration with ripgrep (rg) for rapid searching and filtering of stored memory via summary, tag, or full-text keywords.

• Seamless 'save-and-resume' workflow, allowing agents to log the current state of a task, identified blocking issues, and future action items.

• Standardization of memory format to ensure all entries are machine-readable and easily traversable by the agent's file-system operations.

• Use this skill whenever you discover non-obvious patterns or solutions during debugging to avoid 'forgetting' critical logic.

• Prioritize the summary field in the YAML frontmatter as the primary decision point for the agent when searching through existing notes.

• Adhere to kebab-case naming conventions for files and directories to keep the memory space organized and maintainable.

• Regularly clean up or consolidate memories to avoid fragmentation and ensure the agent only retrieves high-quality, relevant data.

• When performing search operations, always utilize the --no-ignore and --hidden flags with ripgrep to ensure the agent correctly traverses the memory directory despite global gitignore settings.