Project Memory (aim-memory)

Memory is the cognitive backbone of AI-native development. Because LLMs are stateless by default, project context must be stored deterministically in the codebase.

1. Directory Structure

Memory is organized under the .ai-context/ directory:

• .ai-context/config.json: The single source of truth for project metadata, stack, and commands.
• .ai-context/skills/: Holds the active AIM skills.
• .ai-context/docs/: Directory for Markdown documentation files.

2. Tasks & Progress (Task Mode)

During execution, use planning tools to log tasks:

• Maintain a structured task list in your brain directory (task.md).
• Mark tasks with [ ] (todo), [/] (in-progress), and [x] (completed).
• Use Conventional Commits (feat:, fix:, docs:, refactor:) when finalizing task items.

3. Documentation & Cross-Referencing

Keep documentation clean, modular, and cross-referenced:

• Task Mentions: Refer to tasks in files or commit messages using @task-<id>.
• Doc Mentions: Link to Markdown docs using @doc/<path>.
• Code Line Links: Reference specific lines using @doc/<path>:12 or ranges using @doc/<path>:10-25.
• Anchors: Reference markdown sections using @doc/<path>#section-slug.

4. Session memory policy

At the end of a session or task:

• Evaluate whether any new design decision, architecture pattern, user preference, or recurring error should be saved as persistent memory.
• Store project-specific rules in local docs or .ai-context/config.json.
• Store global user preferences in global files so they carry across sessions.
• Default to saving durable rules automatically without waiting for explicit user instructions.
• Never duplicate full documentation code blocks into memory; summarize and reference the file path instead.