feat(recall): entire recall — checkpoint history as trust-ranked associative memory

.gitignore

@@ -71,3 +71,7 @@ tmp/
 entire-external-cp-test/
 /skills/
 /git-remote-entire
+
+# entire recall: derived memory rebuilt from checkpoints, never committed
+.entire/recall/
+recall/target/

CLAUDE.md

@@ -156,7 +156,18 @@ and the inferred one is the common path.
 Experimental commands (gated by the build-time visibility flag above — visible
 and grouped under "Experimental commands:" in developer/nightly builds, hidden
 in stable releases, always runnable): `tokens`, `import`, `review`,
-`investigate`, `blame`, `why`, `experts`, `runner`, and `checkpoint policy`.
+`investigate`, `blame`, `why`, `experts`, `recall`, `runner`, and `checkpoint policy`.
+`recall` (`recall_cmd.go`, `recall_collect.go`) is a shim over the Rust crate in
+`recall/`: `recall ingest` walks trailered commits, reads each checkpoint's
+transcript and diff through the persistent store, and pipes JSON to the
+`recall` binary; `recall <question>` renders the binary's ranked hits. Every
+ranking decision lives in the crate — see `docs/recall/`. The shim never drops
+a checkpoint it cannot read: the commit record is ingested with the missing
+fields declared in `unavailable`, the crate answers `unverifiable` for any
+check that needed one of them, and every answer opens with a coverage line
+that says `complete` only when nothing was missing (`recall/README.md`,
+"Privacy boundary"). `--no-transcripts` never opens a transcript;
+`recall/scripts/verify-offline.sh` proves the binary makes no network calls.
 `tokens` is also advertised through `entire labs`.
 
 Top-level lifecycle and standalone commands: `enable`, `disable`, `status`,

cmd/entire/cli/agent_help_cmd.go

@@ -137,11 +137,15 @@ var agentHelpClassification = map[string]agentHelpFacts{
 	// ---- Unlisted: real commands, just not the default view. ---------------
 	"activity": {agentHelpAudienceReadOnly, false},
 	"blame":    {agentHelpAudienceReadOnly, false},
-	"experts":  {agentHelpAudienceReadOnly, false},
-	"labs":     {agentHelpAudienceReadOnly, false},
-	"recap":    {agentHelpAudienceReadOnly, false},
-	"version":  {agentHelpAudienceReadOnly, false},
-	"tokens":   {agentHelpAudienceReadOnly, false},
+	// recall's bare form only reads, but `recall ingest` rebuilds derived
+	// state under .entire/recall, so the group cannot claim read-only.
+	"recall":        {agentHelpAudienceTaskDriven, false},
+	"recall ingest": {agentHelpAudienceTaskDriven, false},
+	"experts":       {agentHelpAudienceReadOnly, false},
+	"labs":          {agentHelpAudienceReadOnly, false},
+	"recap":         {agentHelpAudienceReadOnly, false},
+	"version":       {agentHelpAudienceReadOnly, false},
+	"tokens":        {agentHelpAudienceReadOnly, false},
 	// tokens is read-only, which is a claim about its children too — classified
 	// rather than assumed, so a future mutating child cannot inherit the claim.
 	"tokens profile": {agentHelpAudienceReadOnly, false},

cmd/entire/cli/experimental_wiring_test.go

@@ -12,7 +12,7 @@ import (
 // token of Use).
 var experimentalRootCommands = []string{
 	"tokens", "import", "review", "investigate",
-	"blame", "why", "experts", "runner",
+	"blame", "why", "experts", "runner", "recall",
 }
 
 // withVisible sets the experimental visibility flag for the test and restores

cmd/entire/cli/explain_export_test.go

@@ -643,7 +643,6 @@ type stubCommittedReader struct {
 	err      error                              // err returned for indexes not in contents
 }
 
-//nolint:unparam // test stub; signature matches CheckpointReader.Read.
 func (s *stubCommittedReader) Read(_ context.Context, _ id.CheckpointID) (*checkpoint.CheckpointSummary, error) {
 	return s.summary, nil
 }

cmd/entire/cli/labs.go

@@ -60,6 +60,11 @@ var experimentalCommands = []experimentalCommandInfo{
 		Invocation:  "entire experts",
 		Summary:     "Show agent, skill, and tool provenance for files or topics",
 	},
+	{
+		CommandPath: []string{"recall"},
+		Invocation:  "entire recall",
+		Summary:     "Ask checkpoint history what was decided and why",
+	},
 }
 
 func newLabsCmd() *cobra.Command {
@@ -108,6 +113,7 @@ Try:
   entire blame --help
   entire why --help
   entire experts --help
+  entire recall --help
 `
 }