← ClaudeAtlas

session-checkpointlisted

Write a small, timestamped snapshot of current state to .memory/checkpoints/ periodically throughout a session (not just at the end), so a session that ends abruptly — credit exhaustion, crash, closed terminal — has a recent, resumable state to bootstrap from. Use after completing a meaningful unit of work, before starting something risky or long-running, and periodically during extended sessions — not only when the user asks to "save" or "checkpoint."
NITISH-R-G/agent-memory-protocol · ★ 1 · AI & Automation · score 69
Install: claude install-skill NITISH-R-G/agent-memory-protocol
# Session Checkpoint **Why this exists as a distinct skill from `memory-capture`**: capture writes specific events as they happen; checkpoint writes the *overall current state* on a cadence, as a safety net independent of whether every individual event got captured. Per [ARCHITECTURE.md](../../ARCHITECTURE.md)'s Layer 3, this is the concrete implementation of the ML Mastery guide's "Execution Checkpointing" pattern, adapted from graph-workflow state to project-memory state. ## What a checkpoint contains Small and current, not a full re-summary of the project: ```markdown # Checkpoint: 2026-08-01T14:30:00 ## Currently doing Refactoring auth middleware to extract TokenService (see decisions/0012-token-service.md) ## Just finished Extracted refresh logic into TokenService; existing tests pass ## Next step Write new unit tests for TokenService directly (not just via middleware integration tests) ## Open threads - Haven't decided whether TokenService should own the refresh interval config or receive it injected - Noticed but didn't fix: middleware still has an unrelated logging concern mixed in ``` Four sections, consistently: currently doing, just finished, next step, open threads. This is deliberately small — a checkpoint should take seconds to write, which is what makes writing it *frequently* actually sustainable. ## When to write one - After completing any meaningful unit of work (not every single edit — a completed sub-task, not every file save) - Before starting