Claude /commands Episode 3 — /init, /resume, /exit: The Session Lifecycle Trio
The third episode in the Claude /commands series focuses on the commands that govern the session lifecycle — the bookends and midpoint of every Claude Code working session. Understanding /init, /resume, and /exit is not just useful for individual developers — it is foundational knowledge for anyone building agentic pipelines on top of Claude Code, where session state management becomes a first-class engineering concern rather than an afterthought.
/init — Generating CLAUDE.md from Codebase Scan
The /init command triggers a codebase scan that produces a CLAUDE.md file — the primary mechanism through which Claude Code retains persistent, project-specific context across sessions. When /init runs, it reads directory structure, language and framework detection signals, configuration files like package.json, pyproject.toml, and Cargo.toml, existing README content, and any detectable architectural patterns such as monorepo layouts or microservice boundaries.
The output structure of the generated CLAUDE.md covers project purpose, technology stack, build and test commands, key file locations, and conventions that Claude should honour. The architectural significance is that CLAUDE.md functions as a machine-readable project brief. Every subsequent session starts with this context pre-loaded, meaning Claude does not need to re-discover the codebase from scratch. For large repositories, this is the difference between a useful first interaction and three turns of orientation questions before any productive work begins.
/resume — Session State Preservation and the Conversation Journal
Claude Code maintains a conversation journal — a structured record of session interactions, decisions made, files modified, and tasks in progress. The /resume command reads this journal and reconstructs the working context of the most recent session. This is not a simple chat history replay. The journal captures semantic state: what problem was being solved, what approach was being taken, and what was left incomplete.
For enterprise developers working across multiple concurrent projects or returning to a complex debugging session after a context switch, /resume eliminates the cognitive overhead of re-establishing context manually. For pipeline architects, the journal mechanism exposes a persistence layer that can be integrated with external orchestration systems — if an agentic pipeline needs to pause a Claude Code session and resume it later, the journal provides the state vehicle for that handoff.
/exit — Graceful Termination and What Gets Persisted
The /exit command performs a graceful session termination. Critically, it triggers a journal flush — ensuring that the current session state is written durably before the process terminates. Without an explicit /exit, abrupt termination may leave the journal in a partial state, degrading the quality of the next /resume. For automated pipelines, this means /exit should be called programmatically at the end of every agent execution. What gets persisted at /exit: the conversation journal entry, any updates to CLAUDE.md made during the session, and any memory entries written via explicit memory operations.
Key Takeaways
/initgenerates a persistentCLAUDE.mdfrom codebase scanning — run it once per project, update it when the architecture changes significantly/resumerestores semantic session context from the conversation journal, not just chat history/exitmust be called explicitly in automated pipelines to ensure clean journal persistence- Treat
CLAUDE.mdand the journal as first-class engineering artifacts, subject to version control and review - Understanding session lifecycle is prerequisite knowledge for building reliable agentic pipelines on Claude Code
