Agent Session Continuity

The problem

AI coding agents reset their memory every time you close the terminal or start a new session. You waste 15 to 30 minutes re-explaining architecture, pointing to modified files, and re-stating goals every time you sit back down to code.

What it does

• Reads and updates a structured session journal in .claude/, .cursor/, or .codex/ directories.
• Tracks Architecture Decision Records (ADR) to ensure agents follow past technical choices.
• Generates machine-readable state.json files to track task progress and blockers across restarts.
• Automatically summarizes the last session status and triggers the next logical step upon startup.

Frameworks & tools

Designed for Claude Code, Cursor, and Codex CLI. Works with any project using ADR for decision logging.

Why this beats prompting it yourself

Manually telling an agent to "remember" rarely works because they lack a systematic way to store and retrieve state. This skill provides a mandatory protocol for writing persistent logs, ensuring the agent retrieves context before you even ask it to.

Use cases

• Resuming a complex feature implementation after a weekend break.
• Switching between multiple projects without manually re-orienting the agent each time.
• Maintaining a consistent architectural style by forcing the agent to check ADRs.

Known limitations

The agent must be allowed to write to the local filesystem to maintain the journal. Very old journals (2+ weeks) may require a manual git log check to sync state.