Crate doover_core 

doover-core — snapshot, journal, and undo primitives for AI agent shell actions.

Module map (built strictly test-first; see doover-implementation-plan.md):

• registry (step 1): reversibility classification of commands
• parser (step 2): bash parsing + affected-path scope resolution
• snapshot (step 3): content-addressed CoW snapshot store
• journal (step 4): SQLite action journal
• hooks (step 5): harness adapters (Claude Code first)
• undo (step 6): restore engine with conflict detection

Modules

hooks

Hook engine: the composition point where harness events become protected actions. handle_pre = parse → resolve scope → snapshot (ALWAYS under limits) → journal pending; handle_post = correlate by tool_use_id → completed. Contract facts baked in from the live capture (fixtures README): the harness sends the session’s live cwd per call, there are no exit codes, and failed commands never emit a post event.

inspect

Read-only inspection: compare a recorded manifest against the live filesystem, per entry (step 8, doover diff).

journal

Append-only action journal (SQLite, WAL).

maintenance

Maintenance / garbage collection (step 7).

redact

Display-time secret redaction for journaled commands (step 8).

registry

Reversibility registry: classifies commands and shell constructs by effect, affected-path scope, and undo strategy. Data lives in registry/*.yaml (CC0); a user overlay directory can add rules or upgrade severity, but a shipped destructive classification can never be silently downgraded.

resolver

Bash command parsing and affected-path scope resolution.

snapshot

Content-addressed snapshot store.

undo

Undo/redo engine (step 6): the user-facing payoff of everything upstream.

Constants

VERSION

Crate version, single source of truth for the CLI --version output.