fix(gmail): remove inbox-only sync restriction to enable sent-mail re… (#1902)

src/openhuman/composio/providers/gmail/provider.rs

@@ -36,6 +36,20 @@ use crate::openhuman::composio::providers::{
 const ACTION_GET_PROFILE: &str = "GMAIL_GET_PROFILE";
 const ACTION_FETCH_EMAILS: &str = "GMAIL_FETCH_EMAILS";
 
+/// Base Gmail search query used on every sync pass.
+///
+/// Excludes spam and trash but intentionally does NOT restrict to `in:inbox` —
+/// that restriction (issue #1713) prevented sent emails from ever being ingested.
+/// Exported `pub(super)` so `tests.rs` can assert against the canonical value
+/// rather than a duplicated literal.
+pub(super) const BASE_QUERY: &str = "-in:spam -in:trash";
+
+/// Gmail search query strings that retrieve sent mail.
+///
+/// Any of these can be passed as the `query` parameter to `GMAIL_FETCH_EMAILS`
+/// to fetch outbound messages. Exported `pub(super)` for use in regression tests.
+pub(super) const SENT_QUERIES: &[&str] = &["from:me", "label:SENT", "in:sent"];
+
 /// Page size per API call. Kept moderate so each call is fast and we
 /// get frequent checkpoints for the daily budget.
 const PAGE_SIZE: u32 = 25;
@@ -283,7 +297,14 @@ impl ComposioProvider for GmailProvider {
             // so same-day re-ticks do not re-fetch a whole day's
             // window every time. Fall back to the day filter only when
             // the cursor cannot be parsed as a timestamp.
-            let mut query = "in:inbox -in:spam -in:trash".to_string();
+            //
+            // NOTE: We intentionally do NOT restrict to `in:inbox` here.
+            // The original query `in:inbox -in:spam -in:trash` meant sent
+            // emails (label:SENT) were never fetched and therefore the
+            // agent could not answer questions about outbound mail (issue #1713).
+            // Removing `in:inbox` lets Gmail return both inbox and sent
+            // messages while still excluding spam and trash.
+            let mut query = BASE_QUERY.to_string();
             if let Some(ref cursor) = state.cursor {
                 if let Some(epoch_filter) = sync::cursor_to_gmail_after_epoch_filter(cursor) {
                     query.push_str(&format!(" after:{epoch_filter}"));

src/openhuman/composio/providers/gmail/tests.rs

@@ -1,5 +1,6 @@
 //! Unit tests for the Gmail provider.
 
+use super::provider::{BASE_QUERY, SENT_QUERIES};
 use super::sync::{
     cursor_to_gmail_after_epoch_filter, cursor_to_gmail_after_filter, extract_messages,
     extract_page_token, now_ms, parse_cursor_to_epoch_secs,
@@ -238,3 +239,63 @@ fn extract_messages_handles_deep_nesting() {
 // live `ComposioClient` (HTTP) plus the global `MemoryClient` singleton.
 // Those go through the integration test suite. Here we just lock in
 // the provider's identity surface and helpers.
+
+// ── Regression tests for issue #1713: sent-mail retrieval ───────────────────
+//
+// Before the fix the sync query was `in:inbox -in:spam -in:trash`, which meant
+// sent emails (label:SENT) were never fetched or ingested into the memory tree.
+// The fix removes `in:inbox` so both inbox and sent mail are fetched.
+
+/// Guard: the provider source must NOT contain the `in:inbox` restriction.
+/// If this test fails, someone reintroduced the inbox-only query that caused
+/// issue #1713.
+#[test]
+fn provider_source_does_not_restrict_to_inbox() {
+    let source = include_str!("provider.rs");
+    // The old restriction started with `"in:inbox`. The double-quote is part
+    // of the Rust string literal, so this catches the exact regression pattern.
+    assert!(
+        !source.contains("\"in:inbox"),
+        "provider.rs sync query must NOT start with 'in:inbox' — this \
+         restriction prevents sent emails from being ingested (issue #1713). \
+         Use '-in:spam -in:trash' to allow both inbox and sent mail."
+    );
+}
+
+/// The base sync query must exclude spam and trash but NOT restrict to inbox.
+/// Asserts against the canonical `BASE_QUERY` constant from provider.rs so
+/// any change to the production value is caught immediately.
+#[test]
+fn sync_base_query_excludes_spam_and_trash_without_inbox_restriction() {
+    assert!(
+        BASE_QUERY.contains("-in:spam"),
+        "BASE_QUERY must exclude spam (got: {BASE_QUERY:?})"
+    );
+    assert!(
+        BASE_QUERY.contains("-in:trash"),
+        "BASE_QUERY must exclude trash (got: {BASE_QUERY:?})"
+    );
+    assert!(
+        !BASE_QUERY.contains("in:inbox"),
+        "BASE_QUERY must NOT restrict to inbox — omitting this allows sent \
+         mail to be fetched and ingested (issue #1713). Got: {BASE_QUERY:?}"
+    );
+}
+
+/// Sent-mail query strings must be non-empty and must not restrict to inbox.
+/// Iterates `SENT_QUERIES` from provider.rs — the canonical list of query
+/// strings a user or agent can pass to GMAIL_FETCH_EMAILS for sent mail.
+#[test]
+fn sent_mail_query_strings_are_well_formed() {
+    assert!(!SENT_QUERIES.is_empty(), "SENT_QUERIES must not be empty");
+    for q in SENT_QUERIES {
+        assert!(
+            !q.is_empty(),
+            "sent-mail query must not be empty, got empty string in SENT_QUERIES"
+        );
+        assert!(
+            !q.starts_with("in:inbox"),
+            "sent-mail query '{q}' must not restrict to inbox"
+        );
+    }
+}

src/openhuman/tools/impl/network/composio.rs

@@ -499,7 +499,10 @@ impl Tool for ComposioTool {
     fn description(&self) -> &str {
         "Execute actions on 1000+ apps via Composio (Gmail, Notion, GitHub, Slack, etc.). \
          Use action='list' to see available actions, action='execute' with action_name/tool_slug, params, and optional connected_account_id, \
-         or action='connect' with app/auth_config_id to get OAuth URL."
+         or action='connect' with app/auth_config_id to get OAuth URL. \
+         For Gmail: GMAIL_FETCH_EMAILS supports standard Gmail search syntax in the 'query' param — \
+         use query='from:me' or query='label:SENT' to retrieve sent emails, query='label:INBOX' for inbox, \
+         query='is:unread' for unread mail, etc. Sent mail is synced and searchable."
     }
 
     fn parameters_schema(&self) -> serde_json::Value {