fix(orchestrator): prefer live integrations over memory_tree for inbox/doc queries (#1731)

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: Rafael Figuereo <rafaelfiguereod@gmail.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>

Cargo.toml

@@ -45,7 +45,7 @@ env_logger = "0.11"
 base64 = "0.22"
 aes-gcm = "0.10"
 argon2 = "0.5"
-rand = "0.9"
+rand = "0.10"
 dirs = "5"
 sha2 = "0.10"
 hmac = "0.12"
@@ -110,7 +110,7 @@ tower = { version = "0.5", default-features = false }
 opentelemetry = { version = "0.31", default-features = false, features = ["trace", "metrics"] }
 opentelemetry_sdk = { version = "0.31", default-features = false, features = ["trace", "metrics"] }
 opentelemetry-otlp = { version = "0.31", default-features = false, features = ["trace", "metrics", "http-proto", "reqwest-client", "reqwest-rustls-webpki-roots"] }
-sentry = { version = "0.47.0", default-features = false, features = ["backtrace", "contexts", "panic", "tracing", "debug-images", "reqwest", "rustls", "test"] }
+sentry = { version = "0.47.0", default-features = false, features = ["backtrace", "contexts", "panic", "tracing", "debug-images", "reqwest", "rustls"] }
 tokio-stream = { version = "0.1.18", features = ["full"] }
 url = "2"
 socketioxide = { version = "0.15", features = ["extensions"] }

scripts/mock-api/state.mjs

@@ -3,6 +3,7 @@ import crypto from "node:crypto";
 export const DEFAULT_PORT = 18473;
 export const MOCK_JWT = "e2e-mock-jwt-token";
 export const MAX_PORT_RETRY_ATTEMPTS = 10;
+export const MAX_MOCK_DELAY_MS = 30_000;
 
 let requestLog = [];
 let mockBehavior = {};
@@ -59,7 +60,8 @@ export function parseBehaviorJson(key, fallback) {
 
 export function getDelayMs(key) {
   const value = Number(mockBehavior[key] || 0);
-  return Number.isFinite(value) && value > 0 ? value : 0;
+  if (!Number.isFinite(value) || value <= 0) return 0;
+  return Math.min(value, MAX_MOCK_DELAY_MS);
 }
 
 export function sleep(ms) {

src/core/auth.rs

@@ -178,10 +178,13 @@ pub async fn rpc_auth_middleware(req: axum::extract::Request, next: Next) -> Res
 ///
 /// Uses `rand::rng()` (thread-local, OS-seeded CSPRNG) introduced in rand 0.9.
 fn generate_token() -> String {
-    use rand::RngCore as _;
+    use rand::RngExt as _;
+    log::trace!("[auth] generate_token: start (32 bytes)");
     let mut bytes = [0u8; 32];
-    rand::rng().fill_bytes(&mut bytes);
-    hex::encode(bytes)
+    rand::rng().fill(&mut bytes);
+    let token = hex::encode(bytes);
+    log::trace!("[auth] generate_token: complete (64 hex chars)");
+    token
 }
 
 /// Write `token` to `path` with owner-only read+write permissions on Unix.

src/openhuman/agent/agents/orchestrator/prompt.md

@@ -17,19 +17,23 @@ Follow this sequence for every user message:
 1. **Can I answer directly without tools?**
    - Yes: reply directly (small talk, simple Q&A, basic factual answers).
    - No: continue.
-2. **Can I solve this with direct tools?**
-   - Yes: use direct tools first (`current_time`, `cron_*`, `memory_*`, `composio_list_connections`, etc.).
+2. **Does the request name (or imply) a connected external service?**
+   - Words like "email/inbox/gmail", "calendar", "notion doc", "drive file", "slack/whatsapp/telegram message", "linear ticket", "send to X", "check X", etc. mean the user wants the **live** service.
+   - Find the matching toolkit in the **Connected Integrations** section and call `delegate_to_integrations_agent` with that `toolkit`.
+   - **Do this even if `memory_tree` could plausibly answer.** The user wants the live source of truth, not a stale summary. Use `memory_tree` only when the user explicitly asks about historical/ingested context (e.g. "what did we discuss last month", "summarise my recent activity") or when a live lookup just failed.
+   - If the relevant toolkit is not in **Connected Integrations**, tell the user to connect it via Settings → Connections → [Service] (see "Connecting external services" below). Do **not** silently fall back to `memory_tree`.
+3. **Can I solve this with direct tools?**
+   - Yes: use direct tools (`current_time`, `cron_*`, `memory_*`, `composio_list_connections`, etc.).
    - No: continue.
-3. **Does this need specialised execution?**
-   - If external SaaS integration work is required, use `delegate_to_integrations_agent` with `toolkit` set to the relevant slug (see the **Connected Integrations** section for the current list).
+4. **Does this need other specialised execution?**
    - If code writing/execution/debugging is required, use `delegate_run_code`.
    - If web/doc crawling is required, use `delegate_researcher`.
    - If complex multi-step decomposition is required, use `delegate_plan`.
    - If code review is requested, use `delegate_critic`.
    - If memory archiving or distillation is required, use `delegate_archivist`.
-4. **After delegation**, summarise results clearly and concisely.
+5. **After delegation**, summarise results clearly and concisely.
 
-Default bias: **do not spawn a sub-agent when a direct response or direct tool call is sufficient**. Use `spawn_worker_thread` for long tasks that need their own thread.
+Default bias: **do not spawn a sub-agent when a direct response or direct tool call is sufficient** — but a live external-service request is *not* something to answer from memory, it requires the integration. Use `spawn_worker_thread` for long tasks that need their own thread.
 
 ## Rules
 
@@ -108,19 +112,33 @@ checking notion
 
 "Q2 roadmap" — 3 bullets: ship auth, cut v0.4, hire designer
 ```
+(`delegate_to_integrations_agent` with `toolkit: "notion"`. The user wants the live doc, not a memory summary.)
+
+User: any new emails from alice today?
+→
+```text
+checking gmail
+
+one, 2pm: "lunch friday?", wants to grab food, no agenda
+```
+(`delegate_to_integrations_agent` with `toolkit: "gmail"`. Do **not** start with `memory_tree`; the user is asking about live inbox state.)
 
 Short answers can skip the ack:
 
 User: what time is it?
 → `7:31pm`
 
-## Memory tree retrieval
+## Memory tree retrieval (historical context only)
 
-Use `memory_tree` with a `mode` argument to query the user's ingested email/chat/document history:
+`memory_tree` queries the user's **already-ingested** email/chat/document history. It is a retrospective index, **not** a live API for connected services. If the user is asking what's in their inbox / calendar / docs *right now*, use `delegate_to_integrations_agent` instead (step 2 of the decision tree).
 
-- `mode: "search_entities"` — resolve a name to a canonical id (e.g. "alice" → `email:alice@example.com`). ALWAYS call this first when the user mentions someone by name.
+Reach for `memory_tree` when the user asks about prior context that's already been summarised — "what did Alice and I discuss last month", "summarise my recent activity", "remind me what we decided on Q2 roadmap" — or when a live integration call has just failed and a stale answer is still useful.
+
+Modes:
+
+- `mode: "search_entities"` — resolve a name to a canonical id (e.g. "alice" → `email:alice@example.com`). Call this first when the user mentions someone by name *and* you've decided memory_tree is the right tool.
 - `mode: "query_topic"` — all cross-source mentions of an `entity_id` from `search_entities`.
-- `mode: "query_source"` — filter by `source_kind` (chat/email/document) and `time_window_days`. Use for "in my email last week…" intents.
+- `mode: "query_source"` — filter by `source_kind` (chat/email/document) and `time_window_days`. Use for retrospective "in my email last week…" intents — **not** for live "check my inbox" intents.
 - `mode: "query_global"` — cross-source daily digest over `time_window_days` (7-day digest is pre-loaded into context on session start — only call for a different window or to force refresh).
 - `mode: "drill_down"` — expand a coarse `node_id` summary one level.
 - `mode: "fetch_leaves"` — pull raw `chunk_ids` for citation.

src/openhuman/agent/agents/orchestrator/prompt.rs

@@ -158,8 +158,12 @@ mod tests {
         let body = build(&ctx_with(&[])).unwrap();
         assert!(body.contains("## Delegation Decision Tree (Direct-First)"));
         assert!(body.contains(
-            "Default bias: **do not spawn a sub-agent when a direct response or direct tool call is sufficient**."
+            "Default bias: **do not spawn a sub-agent when a direct response or direct tool call is sufficient**"
         ));
+        // Step 2 of the decision tree now explicitly routes live external-service
+        // requests to `delegate_to_integrations_agent` rather than `memory_tree`.
+        assert!(body.contains("Does the request name (or imply) a connected external service?"));
+        assert!(body.contains("Do this even if `memory_tree` could plausibly answer"));
     }
 
     #[test]

src/openhuman/memory/tree/tree_source/registry.rs

@@ -107,9 +107,8 @@ fn new_tree_id(kind: TreeKind) -> String {
 /// sized for uniqueness across the file-system and Obsidian wikilink
 /// namespaces.
 pub fn new_summary_id(level: u32) -> String {
-    use rand::Rng;
     let ms = chrono::Utc::now().timestamp_millis() as u64;
-    let rand_tail: u32 = rand::thread_rng().gen();
+    let rand_tail: u32 = rand::random();
     format!("summary:{:013}:L{}-{:08x}", ms, level, rand_tail)
 }
 

src/openhuman/security/pairing.rs

@@ -209,9 +209,9 @@ fn generate_code() -> String {
 /// on macOS). The 32 random bytes (256 bits) are hex-encoded for a
 /// 64-character token, providing 256 bits of entropy.
 fn generate_token() -> String {
-    use rand::RngCore;
+    use rand::RngExt as _;
     let mut bytes = [0u8; 32];
-    rand::rng().fill_bytes(&mut bytes);
+    rand::rng().fill(&mut bytes);
     format!("zc_{}", hex::encode(bytes))
 }
 

src/openhuman/tools/impl/computer/human_path.rs

@@ -1,4 +1,4 @@
-use rand::Rng;
+use rand::RngExt;
 use std::f64::consts::TAU;
 
 #[derive(Debug, Clone, Copy)]
@@ -25,7 +25,7 @@ impl Default for HumanPathOptions {
 }
 
 /// Returns `(x, y, dwell_ms)` steps for a humanized cursor path.
-pub fn human_path<R: Rng>(
+pub fn human_path<R: RngExt>(
     start: (i32, i32),
     end: (i32, i32),
     opts: &HumanPathOptions,
@@ -95,7 +95,7 @@ fn cubic_bezier(
     )
 }
 
-fn dwell_ms<R: Rng>(opts: &HumanPathOptions, rng: &mut R) -> u64 {
+fn dwell_ms<R: RngExt>(opts: &HumanPathOptions, rng: &mut R) -> u64 {
     let mean = finite_or_default(opts.mean_step_ms, HumanPathOptions::default().mean_step_ms);
     let stddev = finite_or_default(
         opts.stddev_step_ms,
@@ -119,7 +119,7 @@ fn finite_or_default(value: f64, default: f64) -> f64 {
     }
 }
 
-fn sample_normal<R: Rng>(mean: f64, stddev: f64, rng: &mut R) -> f64 {
+fn sample_normal<R: RngExt + RngExt>(mean: f64, stddev: f64, rng: &mut R) -> f64 {
     if stddev <= 0.0 {
         return mean;
     }
@@ -127,7 +127,7 @@ fn sample_normal<R: Rng>(mean: f64, stddev: f64, rng: &mut R) -> f64 {
         .random::<f64>()
         .clamp(f64::MIN_POSITIVE, 1.0 - f64::EPSILON);
     let u2 = rng.random::<f64>();
-    let z0 = (-2.0 * u1.ln()).sqrt() * (TAU * u2).cos();
+    let z0 = (-2.0_f64 * u1.ln()).sqrt() * (TAU * u2).cos();
     mean + z0 * stddev
 }