checkpoint-protocol

Cross-skill session persistence is achieved by writing a standardized JSON checkpoint file to the project's .checkpoints directory. This file captures phase, step, status, and context to resume multi-session workflows deterministically after a reload.

A JSON checkpoint format standardizes cross-skill workflow coordination by defining a structured contract with fields like protocol_version, project, progress_table, blockers, and next_actions. This schema allows different skills to read, resume, and coordinate multi-phase progress.

To resume a multi-phase project after losing context, read the checkpoint JSON file stored in the project's .checkpoints directory. The file's progress_table and context_primer fields provide the exact state needed to restart the workflow deterministically where it left off.

Yes, checkpoint files enable cross-skill reads by standardizing progress data in a shared JSON format within the .checkpoints directory. Any skill can read the progress_table and optional skill_state fields to pick up the workflow from the current phase.

Migrating older state representations to a JSON checkpoint model requires mapping existing session data to the standardized checkpoint fields like phase, step, status, and progress_table. The protocol provides migration guidelines to structure older data into the deterministic JSON format.

No external dependencies are required to enable checkpoint persistence. The protocol operates independently by managing JSON files directly in the {project-dir}/.checkpoints folder, allowing any project to implement state saving without external libraries.