Run Composio normalisation and storage on a driver that is not TinyCortex (#18 §B3)

Cargo.toml

@@ -1,6 +1,7 @@
 [workspace]
-members = [".", "api", "core", "adapters/tinycortex", "adapters/remote", "conformance"]
-default-members = [".", "api", "core", "adapters/tinycortex", "adapters/remote", "conformance"]
+# `sync` is the engine-neutral Composio normalisers (issue #18 §B3).
+members = [".", "api", "core", "sync", "adapters/tinycortex", "adapters/remote", "conformance"]
+default-members = [".", "api", "core", "sync", "adapters/tinycortex", "adapters/remote", "conformance"]
 # `vendor/` holds engine submodules (tinycortex, tinybus, tinyagents), each of
 # which is its own workspace with its own lockfile. Same exclusion
 # `vendor/tinycortex` uses for its own nested vendor directory.
@@ -79,6 +80,9 @@ tinymemory-api = { path = "api", features = ["test-support"] }
 # integration tests. A dev-dependency only: the facade must not carry a test
 # harness into a consumer's dependency graph.
 tinymemory-conformance = { path = "conformance" }
+# The extracted Composio normalisers, for the integration test that runs them
+# against a driver that is not TinyCortex (issue #18 §B3).
+tinymemory-sync = { path = "sync" }
 
 [features]
 default = []

conformance/src/suite/mod.rs

@@ -601,7 +601,7 @@ pub async fn assert_awkward_content_round_trips(provider: &dyn MemoryProvider) {
         ("large", "x".repeat(64 * 1024)),
         ("newlines", "a\nb\r\nc\0d".to_string()),
     ];
-    let mut accepted = 0usize;
+    let mut accepted: Vec<&str> = Vec::new();
     for (key, content) in &cases {
         // A driver may refuse a shape outright — `MemoryCore::store` documents
         // `Invalid` "for caller input the driver rejects", and the TinyCortex
@@ -629,7 +629,7 @@ pub async fn assert_awkward_content_round_trips(provider: &dyn MemoryProvider) {
         {
             continue;
         }
-        accepted += 1;
+        accepted.push(key);
         if let Some(got) = provider
             .get(&ns, key)
             .await
@@ -638,12 +638,17 @@ pub async fn assert_awkward_content_round_trips(provider: &dyn MemoryProvider) {
             assert_eq!(&got.content, content, "{who}: `{key}` content was mangled");
         }
     }
-    // Without this a driver that refused every shape would pass having stored
-    // nothing, which is the vacuous reading of "may refuse".
+    // "May refuse" needs a floor, or a driver that refused everything would pass
+    // having stored nothing. The floor is `unicode` specifically rather than a
+    // count: refusing `empty` is documented validation, and refusing `large` is
+    // a defensible size limit, but refusing ordinary UTF-8 text is a broken
+    // driver — and unicode is the case where mangling actually shows, since
+    // truncation and re-encoding are invisible on ASCII.
     assert!(
-        accepted > 0,
-        "{who}: refused every content shape — unicode, empty, large and \
-         newlines were all rejected, so this assertion proved nothing"
+        accepted.contains(&"unicode"),
+        "{who}: refused ordinary UTF-8 content — accepted {accepted:?}. A driver \
+         may refuse a shape, but not this one; every assertion about content \
+         surviving a round trip rests on it."
     );
     let keys: Vec<&str> = cases.iter().map(|(k, _)| *k).collect();
     cleanup(provider, &ns, &keys).await;

core/Cargo.toml

@@ -16,6 +16,11 @@ readme = "../README.md"
 # The contract. `tinymemory-core` implements and consumes it; the host seam
 # traits (config, event sink, embeddings, chat) live in `tinymemory_api::host`.
 tinymemory-api = { path = "../api" }
+# Composio payload normalisers, extracted out of the engine (issue #18 §B3).
+# They were reached through `tinycortex` until now, which meant a host binding a
+# different engine could not have Composio sync despite none of this code
+# caring which engine is bound.
+tinymemory-sync = { path = "../sync" }
 tinymemory = { path = ".." }
 
 # The default embedded engine. `store/`, `tree/` and `sync/` drive it directly;

core/src/sync/composio/providers/clickup/mod.rs

@@ -17,7 +17,7 @@
 // The payload normalisers moved to tinycortex (they are pure Value
 // transforms, i.e. driver-side). Aliased under the old module name so
 // every `normalization::extract_*` call site below stays unchanged.
-use crate::engine::backend::sync::composio::providers::normalize::clickup as normalization;
+use tinymemory_sync::clickup as normalization;
 mod provider;
 #[cfg(test)]
 mod tests;

core/src/sync/composio/providers/github/mod.rs

@@ -16,7 +16,7 @@
 // The payload normalisers moved to tinycortex (they are pure Value
 // transforms, i.e. driver-side). Aliased under the old module name so
 // every `normalization::extract_*` call site below stays unchanged.
-use crate::engine::backend::sync::composio::providers::normalize::github as normalization;
+use tinymemory_sync::github as normalization;
 mod provider;
 #[cfg(test)]
 mod tests;

core/src/sync/composio/providers/gmail/mod.rs

@@ -1,7 +1,7 @@
 // The Gmail post-processor moved to tinycortex (a pure Value transform, i.e.
 // driver-side). Aliased under the old module name so the single call site in
 // `provider.rs` stays unchanged.
-use crate::engine::backend::sync::composio::providers::normalize::gmail_post_process as post_process;
+use tinymemory_sync::gmail_post_process as post_process;
 mod provider;
 #[cfg(test)]
 mod tests;

core/src/sync/composio/providers/helpers.rs

@@ -1,11 +1,11 @@
 //! Shared helpers for Composio provider implementations.
 //!
 //! `pick_str` used to live here. It is a provider payload normaliser, so it
-//! moved to `crate::engine::backend::sync::composio::providers::normalize::helpers`
+//! moved to `tinymemory_sync::helpers`
 //! and is re-exported from this module's parent. The helpers that remain are
 //! request-building rather than normalisation, and stay host-side.
 
-use crate::engine::backend::sync::composio::providers::normalize::helpers::pick_str;
+use tinymemory_sync::helpers::pick_str;
 
 /// Shallow-merge an `extra` JSON object into a (mutable) action-args
 /// object. Only object-typed extras are merged; non-object `extra`

core/src/sync/composio/providers/linear/mod.rs

@@ -6,7 +6,7 @@
 // The payload normalisers moved to tinycortex (they are pure Value
 // transforms, i.e. driver-side). Aliased under the old module name so
 // every `normalization::extract_*` call site below stays unchanged.
-use crate::engine::backend::sync::composio::providers::normalize::linear as normalization;
+use tinymemory_sync::linear as normalization;
 mod provider;
 #[cfg(test)]
 mod tests;

core/src/sync/composio/providers/mod.rs

@@ -281,11 +281,11 @@ pub(crate) use helpers::{first_array_str, merge_extra};
 // re-exported here so the ~40 in-tree call sites keep resolving unchanged.
 // Note this is deliberately NOT `providers::common::pick_str`, which coerces
 // numbers to strings — see the doc comments on both definitions.
-pub(crate) use crate::engine::backend::sync::composio::providers::normalize::helpers::pick_str;
 pub use registry::{
     all_providers, get_provider, init_default_providers, register_provider, ProviderArc,
 };
 pub use scope_lookup::{curated_scope_for, toolkit_has_scope};
+pub(crate) use tinymemory_sync::helpers::pick_str;
 pub use tool_scope::{classify_unknown, find_curated, toolkit_from_slug, CuratedTool, ToolScope};
 pub use traits::{resolve_sync_interval_secs, sync_interval_env_var, ComposioProvider};
 pub use types::{

core/src/sync/composio/providers/notion/mod.rs

@@ -1,7 +1,7 @@
 // The payload normalisers moved to tinycortex (they are pure Value
 // transforms, i.e. driver-side). Aliased under the old module name so
 // every `normalization::extract_*` call site below stays unchanged.
-use crate::engine::backend::sync::composio::providers::normalize::notion as normalization;
+use tinymemory_sync::notion as normalization;
 mod provider;
 #[cfg(test)]
 mod tests;

core/src/sync/composio/providers/slack/mod.rs

@@ -10,7 +10,7 @@
 // driver-side). Re-exported under the old module name — `pub`, not a plain
 // `use`, because `tests/raw_coverage/memory_threads_raw_coverage_e2e.rs`
 // imports this path directly.
-pub use crate::engine::backend::sync::composio::providers::normalize::slack_post_process as post_process;
+pub use tinymemory_sync::slack_post_process as post_process;
 pub mod types;
 
 mod provider;

scripts/ci/dependency-budget.sh

@@ -38,6 +38,26 @@ printf '%-52s %s\n' "tinymemory-api" "$(count -p tinymemory-api)"
 printf '%-52s %s\n' "tinymemory-tinycortex (default)" "$(count -p tinymemory-tinycortex --no-default-features)"
 printf '%-52s %s\n' "tinymemory-tinycortex --features memory-git" "$(count -p tinymemory-tinycortex --features memory-git)"
 printf '%-52s %s\n' "tinymemory-remote" "$(count -p tinymemory-remote)"
+printf '%-52s %s\n' "tinymemory-sync" "$(count -p tinymemory-sync)"
+
+# The sync crate exists because it has no engine behind it (issue #18 §B3):
+# the Composio normalisers lived inside TinyCortex, and a host binding a
+# different engine could not run them. Nothing else enforces that, and a
+# dependency added two crates away would reintroduce the coupling silently —
+# the build would still be green, and the property would just quietly stop
+# being true.
+sync_engine="$(
+  cargo tree -p tinymemory-sync -e normal --prefix none 2>/dev/null \
+    | grep -Ei '^(tinycortex|rusqlite|libsqlite|tinymemory-core|tinymemory-api)' || true
+)"
+if [ -n "$sync_engine" ]; then
+  echo "tinymemory-sync reached an engine, a store, or the contract:" >&2
+  echo "$sync_engine" >&2
+  echo >&2
+  echo "That crate is the one piece of Composio sync a non-TinyCortex host can" >&2
+  echo "use. A dependency on any of the above puts it back behind an engine." >&2
+  exit 1
+fi
 
 echo
 if [ "$minimal" -gt "$MINIMAL_CEILING" ]; then

sync/Cargo.toml

@@ -0,0 +1,45 @@
+[package]
+name = "tinymemory-sync"
+publish = false
+version = "0.1.0"
+edition = "2021"
+rust-version = "1.96"
+license = "MIT"
+repository = "https://github.com/tinyhumansai/tinymemory"
+description = "Engine-neutral Composio payload normalisers for TinyMemory"
+
+# The whole dependency list, and it is the point of the crate. These are pure
+# `Value -> Value` transforms: no engine, no storage, no network, no async
+# runtime. A dependency added here should have to argue for itself against that
+# sentence (issue #18 §B3).
+[dependencies]
+serde_json = "1"
+# Two logging facades, neither an implementation, both carried over from the
+# engine layout this crate was extracted from: `gmail_post_process` traces
+# through `tracing`, `slack_post_process` through `log`. Preserved rather than
+# unified, because §B3 is a *move* and swapping a facade would change where a
+# host's log lines surface — a behaviour change hiding inside a relocation.
+# Worth reconciling in its own change.
+tracing = "0.1"
+log = "0.4"
+# RFC 2822/3339 date handling for Gmail `Date:` headers.
+#
+# `clock` is on, and it is the one place this crate is not a pure function of
+# its input: `format_email_local_time` renders in `chrono::Local`, so it reads
+# the host's timezone. That is deliberate upstream — the agent presents local
+# times without doing UTC arithmetic, and the raw UTC field is preserved
+# alongside — but it means "pure `Value -> Value`" is true of every normaliser
+# here except that one. Better said out loud than discovered by someone whose
+# output moved when they changed TZ.
+chrono = { version = "0.4", features = ["clock"] }
+
+[lints.rust]
+unsafe_code = "forbid"
+missing_docs = "warn"
+unreachable_pub = "warn"
+
+[lints.clippy]
+all = { level = "warn", priority = -1 }
+unwrap_used = "warn"
+expect_used = "warn"
+panic = "warn"