fix(core): enforce managed plaintext injection policy

Co-authored-by: Codex noreply@openai.com

codex-rs/network-proxy/src/state.rs

@@ -30,6 +30,7 @@ pub struct NetworkProxyConstraints {
     pub allow_upstream_proxy: Option<bool>,
     pub dangerously_allow_non_loopback_proxy: Option<bool>,
     pub dangerously_allow_all_unix_sockets: Option<bool>,
+    pub dangerously_allow_plaintext_credential_injection: Option<bool>,
     pub allowed_domains: Option<Vec<String>>,
     pub allowlist_expansion_enabled: Option<bool>,
     pub denied_domains: Option<Vec<String>>,
@@ -198,6 +199,27 @@ pub fn validate_policy_against_constraints(
         },
     )?;
 
+    if let Some(allow_plaintext_credential_injection) =
+        constraints.dangerously_allow_plaintext_credential_injection
+    {
+        validate(
+            config
+                .network
+                .dangerously_allow_plaintext_credential_injection,
+            move |candidate| {
+                if *candidate && !allow_plaintext_credential_injection {
+                    Err(invalid_value(
+                        "network.dangerously_allow_plaintext_credential_injection",
+                        "true",
+                        "false (disabled by managed config)",
+                    ))
+                } else {
+                    Ok(())
+                }
+            },
+        )?;
+    }
+
     let allow_all_unix_sockets = constraints
         .dangerously_allow_all_unix_sockets
         .unwrap_or(constraints.allow_unix_sockets.is_none());