test(agent): parity test backfill — credential scrub + failure classification (#4468) (#4509)

src/openhuman/agent/harness/credentials.rs

@@ -111,4 +111,53 @@ mod tests {
         let output = scrub_credentials(input);
         assert!(output.contains("api_key: 1234*[REDACTED]"));
     }
+
+    // #4453: bare, unlabelled secrets that show up in env dumps / API responses.
+
+    #[test]
+    fn scrubs_bare_aws_access_key() {
+        let out = scrub_credentials("config dump AKIAIOSFODNN7EXAMPLE trailing text");
+        assert!(
+            !out.contains("AKIAIOSFODNN7EXAMPLE"),
+            "bare AWS access key must be redacted: {out}"
+        );
+        assert!(
+            out.contains("[REDACTED]"),
+            "redaction marker present: {out}"
+        );
+    }
+
+    #[test]
+    fn scrubs_bare_openai_key() {
+        let out = scrub_credentials("response body sk-abcdefghij1234567890ABCDEFGHIJ end");
+        assert!(
+            !out.contains("abcdefghij1234567890ABCDEFGHIJ"),
+            "openai secret body must be redacted: {out}"
+        );
+        assert!(
+            out.contains("sk-"),
+            "the sk- scheme is kept for context: {out}"
+        );
+        assert!(
+            out.contains("[REDACTED]"),
+            "redaction marker present: {out}"
+        );
+    }
+
+    #[test]
+    fn scrubs_space_separated_bearer_token() {
+        let out = scrub_credentials("Authorization: Bearer abcDEF1234567890ghijklmnop done");
+        assert!(
+            !out.contains("abcDEF1234567890ghijklmnop"),
+            "space-separated bearer token must be redacted: {out}"
+        );
+        assert!(
+            out.contains("Bearer"),
+            "the scheme word is kept for context: {out}"
+        );
+        assert!(
+            out.contains("[REDACTED]"),
+            "redaction marker present: {out}"
+        );
+    }
 }

src/openhuman/tinyagents/middleware.rs

@@ -3140,6 +3140,46 @@ impl Middleware<()> for ImageAwareMessageTrimMiddleware {
 #[cfg(test)]
 mod tests {
     use super::*;
+
+    // #4462: image-aware token estimation. A base64 image marker must be priced
+    // at the flat IMAGE_MARKER_TOKEN_COST, not chars/4 of its payload — otherwise
+    // one image reads as millions of tokens and the trimmer evicts everything,
+    // including the system prompt.
+
+    #[test]
+    fn estimate_text_tokens_markerless_is_chars_over_four() {
+        assert_eq!(estimate_text_tokens(&"a".repeat(40)), (40 + 3) / 4);
+        assert_eq!(estimate_text_tokens(""), 0);
+    }
+
+    #[test]
+    fn estimate_text_tokens_prices_image_marker_flat_not_by_length() {
+        let huge = "x".repeat(40_000);
+        let text = format!("[IMAGE:{huge}]");
+        let tokens = estimate_text_tokens(&text);
+        // chars/4 of the payload would be ~10_000; the flat price is 1_200.
+        assert!(
+            tokens >= IMAGE_MARKER_TOKEN_COST,
+            "at least the flat image cost: {tokens}"
+        );
+        assert!(
+            tokens < 2_000,
+            "image priced flat, not by base64 length: {tokens}"
+        );
+    }
+
+    #[test]
+    fn estimate_text_tokens_charges_each_image_marker_once() {
+        let tokens = estimate_text_tokens("[IMAGE:aaaa] and [IMAGE:bbbb]");
+        assert!(
+            tokens >= 2 * IMAGE_MARKER_TOKEN_COST,
+            "two images each priced: {tokens}"
+        );
+        assert!(
+            tokens < 2 * IMAGE_MARKER_TOKEN_COST + 100,
+            "no runaway from the surrounding text: {tokens}"
+        );
+    }
     use serde_json::json;
     use tinyagents::harness::context::{RunConfig, RunContext};
     use tinyagents::harness::model::ModelRequest;

src/openhuman/tinyagents/tests.rs

@@ -684,3 +684,30 @@ fn record_unobserved_turn_usage_gates_on_observed_tokens() {
     assert!(record_unobserved_turn_usage("m", 0, 3, 0, 0.0));
     assert!(record_unobserved_turn_usage("m", 10, 3, 2, 0.5));
 }
+
+#[test]
+fn spawn_and_delegate_tools_are_never_registered_on_subagents() {
+    // #4452: a child run must never be able to register a spawn/delegate tool,
+    // even if the resolved allowlist somehow contains one — the registration
+    // site strips these unconditionally as defense-in-depth.
+    for name in [
+        "spawn_subagent",
+        "spawn_worker_thread",
+        "use_tinyplace",
+        "agent_prepare_context",
+        "delegate_research",
+        "delegate_",
+    ] {
+        assert!(
+            is_subagent_spawn_or_delegate_tool(name),
+            "{name} must be treated as a spawn/delegate tool"
+        );
+    }
+    // Ordinary tools (and near-miss names) must NOT be stripped.
+    for name in ["shell", "read_file", "web_search", "spawn", "subagent"] {
+        assert!(
+            !is_subagent_spawn_or_delegate_tool(name),
+            "{name} is a normal tool and must not be stripped"
+        );
+    }
+}

src/openhuman/tool_status/ops.rs

@@ -388,6 +388,27 @@ mod tests {
         );
     }
 
+    #[test]
+    fn policy_denied_marker_is_denied_and_not_recoverable() {
+        use crate::openhuman::security::POLICY_DENIED_MARKER;
+        let text = format!("{POLICY_DENIED_MARKER} you declined this shell action");
+        assert_eq!(class_of(&text), ToolFailureClass::Denied);
+        // UserDeclined family — never eligible for an auto-retry (#4459).
+        assert!(!classify(&text, false).recoverable);
+    }
+
+    #[test]
+    fn policy_denied_ttl_expiry_is_approval_expired_not_timeout() {
+        use crate::openhuman::security::POLICY_DENIED_MARKER;
+        // A TTL-expiry deny reason literally contains "timed out" — the policy
+        // marker must win over the timeout sniff so it classifies as an expired
+        // approval, NOT an execution Timeout that promises an auto-retry (#4459).
+        let text = format!("{POLICY_DENIED_MARKER} Approval for 'shell' timed out after 600s");
+        assert_eq!(class_of(&text), ToolFailureClass::ApprovalExpired);
+        assert_ne!(class_of(&text), ToolFailureClass::Timeout);
+        assert!(!classify(&text, false).recoverable);
+    }
+
     #[test]
     fn numeric_status_codes_need_word_boundaries() {
         // `403`/`503` embedded in a longer digit run must NOT trip the code