Local State And Snapshot Reference

criv watch --once and live watch publication write the current graph to .criv/state.json. They also keep content-addressed local snapshots for criv query diff. The snapshot store is ignored local data; source control remains the durable history boundary.

Configure the maximum number of distinct local snapshots in criv.toml:

[state]
keep = 20

keep must be a positive integer and defaults to 20. The bound is applied after every successful state publication. Publishing identical content again does not consume another slot, but makes that hash the newest publication. The snapshot named by .criv/latest is always protected.

Inspect retained snapshots newest first:

criv state list
criv state list --format json

Preview or apply pruning with the configured bound:

criv state prune --dry-run
criv state prune
criv state prune --keep 5
criv state prune --keep 5 --format json

--keep is a one-command positive override; it does not rewrite criv.toml. --dry-run selects and reports the same snapshots without changing local files. Corrupt recognized snapshots fail closed and are not deleted automatically.

criv query diff <a> <b> resolves latest and retained hashes from the local store. Other values remain embedded Git-ref lookups of .criv/state.json. Pruning local snapshots does not inspect or alter Git refs.

The lifecycle and compatibility boundary are defined by ADR-0068.